Modeling With NLP*

by Steve Andreas
© 1999

      This article is an outline of some principles and ideas about modeling that my wife, Connirae, and I have been using as a basis for modeling projects in the NLP Comprehensive Master Practitioner training for some years, based on our experience of modeling a number of NLP patterns. I offer it as an additional perspective in the current dialogue about modeling.

General Principles


     A model is only a more-or-less sophisticated metaphor for understanding some part of the world. When physicists describe the behavior of an electron as a “particle,” it leads naturally to some kinds of understandings and predictions, and tends to exclude others. When physicists describe an electron as a “wave,” they discover understandings and applications that are not available to them when thinking of an electron as a particle. What is an electron “really”? Undoubtedly neither a “wave” nor a “particle.” Hopefully someday someone will come up with a new metaphor that comes closer to describing what an electron “really” is, and which yields deeper and more extensive understandings. Some physicists are now using the metaphor of a “string,” which has both particle and wave qualities, and holds forth the possibility of integrating the understandings that have been gained from both the particle and wave models, and perhaps may suggest new understandings. I am not sufficiently educated about contemporary physics to know how useful this new description has been to date.
    Freud’s thinking about feelings and emotions was based on a hydraulic or “plumbing” metaphor (following Descartes’ theory of how the brain worked). He thought of feelings as being fluids that were stored, and if they were pushed down in one area of life they would squirt out somewhere else. Primal therapy, an offshoot, spoke of a “primal pool of pain” that could be “drained” by screaming.
    In contrast, the NLP metaphor is that of an information system that stores information as recorded memories in one or more representational systems, corresponding to the five senses. It is only when these records are activated that feelings result from them (if they are meaningful). If the memory is never activated, no feelings are stimulated. A CD player has lots of records of music, but it is only when the laser beam reads these records that there is music. If we used the Freudian metaphor to describe this, we’d say that the CD is full of music struggling for expression (catharsis). Thinking of a person as an information system makes it clear why catharsis not only doesn’t work but can make many problems worse, or even create new ones. Although the information system metaphor has been much more useful than the Freudian plumbing metaphor, yet another one (not yet discovered) may prove to be even better.

Descriptive Language and Injunctive Language

    Some models are purely descriptive, and do not tell you what to do. Transactional analysis (TA) was an elaborate description that basically recast Freud’s parts model of the id, ego, and superego into contemporary English (child, adult, and parent). However, this description did not provide any methodology or technology for making use of the description. In practice, TA borrowed methods and techniques from other forms of psychotherapy and adapted them to their descriptive framework. Psychiatry’s “Diagnostic and Statistical Manual” has over 700 pages describing how people can have problems, but not a word about what to do to resolve them.
     In contrast, injunctive language tells you what to do to reach an outcome.
A cookbook is injunctive, because each recipe tells you exactly what to do to get a particular result. A recipe specifies:

     1. an outcome (chocolate cake or a well-done roast)
    2. a list of required ingredients (flour, sugar, chocolate, etc., or a particular cut of meat)
    3. a sequential process for mixing and processing those ingredients and how to cook them to get the desired result.

   A cookbook is essentially a list of techniques, and someone can follow the instructions and get the result without any understanding of the processes involved. A number of years ago I listened to an audiotape of someone teaching and demonstrating the Forgiveness Pattern that Connirae and I developed along with the participants in a modeling seminar in 1990. (An article about this process can be found on the NLP Comprehensive web site at: ) The presenter’s theory about how the forgiveness process worked was very complex, but had no resemblance to our understanding of it. Nevertheless, he led the demonstration subject successfully through all the steps of the technique and into the experience of forgiveness. In one sense, what he did is the highest compliment one can pay to a technique–that the instructions are sufficiently detailed and precise that someone can use it without any understanding (or even with an inappropriate understanding) and it will still work.

    “The term science should not be given to anything but the aggregate of the recipes that are always successful”
                                       –Paul Valery

    All of us are surrounded by technology that we use, but do not understand. When we use a cell phone, an automatic transmission, or an antibiotic, most of us don’t have the vaguest idea of the physics or chemistry involved. No human being lives long enough to understand even a small fraction of existing technology, even if s/he spent a lifetime studying it.


     Technology is a specific application of a methodology (whether the methodology is known or understood or not). Bronze age people discovered that when arsenic or tin was added to copper, it made it much harder, but they had no idea why that was so; the methodology came five thousand years later with the understanding of how small amounts of impurities “lock up” layers of atoms so that they don’t slip and deform when shearing force is applied to them.
    A methodology is a more general understanding of how things work, in contrast to a specific recipe or product. Richard Bandler has often said that “NLP is a methodology that leaves behind it a trail of techniques.” Most NLP trainings include a mixture of methodology and technology. Specific techniques (e.g. phobia procedure, change personal history) are taught along with at least part of the methodology (e.g. rep. systems, submodalities, anchoring) that underlie the specific techniques.
   Knowledge of methodology allows the user of technology to adapt it to unique situations in which knowledge of the technique alone would fail. Methodology also makes possible new applications and discoveries, and new ways of accomplishing outcomes that we already have techniques for.
   For example, an engineer who understands the methodology of materials and structures can build a specified building out of a wide variety of materials, utilizing a wide range of structural elements, and predict with mathematical models exactly what size to make everything to achieve a certain strength to resist hazards such as snow load, flood, earthquakes, etc. In contrast, if the same engineer only knew about how to build brick walls, he would only be able to design a narrow range of buildings for a few environments.

Methodology and Technology

     Typically a field develops by a kind of “leap frog” alternation of technology and methodology. Usually some primitive technology, discovered by accident or intuition, starts the process. Then someone looks at several techniques and begins to generalize about them, describing some elements of similarity, usually using a metaphor to describe this understanding. If this generalization is a useful one, typically it indicates other technologies that could be developed using different processes, materials, or outcomes. These new techniques, and the knowledge that is learned as they are applied and tested, in turn suggest other methodologies–other ways of thinking about the technology. Methodology is at a higher, more general (logical) level of generalization than technology.
   An evolving methodology/technology usually has very useful pieces that do not yet appear to fit together. It was a long time before physicists realized how light (and optics) could be described as a part of the electromagnetic radiation spectrum, and they are still seeking an understanding of how gravity and electromagnetics are related.
     In NLP there are a number of different models: anchoring, reframing, rep. systems, strategies, submodalities, “parts,” perceptual positions, etc., and it is seldom clear for instance exactly what submodalities make up a part, where a “part” appears in a strategy sequence, or how reframing can be understood as anchoring. As we make progress in refining our understandings, these relations will become clearer.


       Epistemology is the study of how we know things. My Webster’s unabridged dictionary defines epistemology as “The theory or science that investigates the origin, nature, methods and limits of knowledge.” Every model also has an implicit epistemology, at both the level of technology and methodology.
        Some epistemologies are very simple; they rely on some authority–a person, book or other original source from which the model originates. Most such epistemologies do not have an independent way to test the validity of the model, and typically such methodologies do not develop or change significantly over long periods of time. The basic tenets of astrology, for instance, have not changed much in several thousand years–though practitioners are happy to use an ephemeris created by a very different scientific methodology/technology/epistemology to calculate the positions of the planets at a particular time and place on the earth.
      The scientific method, in contrast, includes a rigorous way of testing and revising methodology, an explicit recognition of the inherent uncertainty in all knowledge, and the testing of this knowledge. As Hans Vaihinger wrote in The Philosophy of “As If,” “Truth is only the most expedient error.” This was echoed by Richard Bandler who said, “Everything we tell you is lies–but they are very useful lies.” This aspect of the epistemology of science essentially says “I don’t care if it’s ‘true;’ I only care if it’s true enough to yield predictions about the world that can be used. The following poem says it even better:

                     Not truth, nor certainty. These I forswore
                     In my novitiate, as young men called
                     To holy orders must abjure the world.
                     “If . . ., then . . .,” this only I assert;
                     And my successes are but pretty chains
                     Linking twin doubts, for it is vain to ask
                     If what I postulate be justified,
                     Or what I prove possess the stamp of fact.

                     Yet bridges stand, and men no longer crawl
                     In two dimensions.And such triumphs stem
                     In no small measure from the power this game,
                     Played with the thrice-attenuated shades
                     Of things, has over their originals.
                     How frail the wand, but how profound the spell!

   In this epistemology, called radical empiricism by Karl Popper, “truth” is a relative, analog, function rather than something absolute, and digital, and it is measured simply by how useful an understanding is.

To summarize, every model has three levels (whether explicit or not):1. Epistemology, a set of methods for discovering and testing understandings.   
2. Methodology, a statement of the understanding itself.   
3. Technology, specific applications of the understanding to accomplish a particular outcome in a particular context.


   The more general a model is, the more it can be applied to a wide range of situations. However, the more general it is, the less information it supplies about specific situations. E=mc2 is understood to apply to the entire universe, but it doesn’t tell you how to make a match or how to build a pump. More limited and specific models can provide more detailed and useful information. One important element of any model is to know the scope of the domain that is usefully described by it. For example, NLP is a wonderful model, but it is not directly useful in designing an automobile engine or telling a doctor how to set a broken bone.


      A new model is created when one realm of experience (e.g. “particle”) is used to describe another (e.g. electron) metaphorically, and then further developed through testing, statements of how to apply and refine this metaphor through mathematics, etc. The initial creative leap is followed by a lot of work to develop the detailed recipes and procedures that make it useful. It took over a hundred and fifty years from Michael Faraday’s discovery of electromagnetic induction (a needle suspended by a thread next to a wire moving in response to a current in the wire)to the giant generators in today’s power plants.
   Once a model has been created it can then be applied to other events within the domain described by the model, or in some cases applied usefully to other domains. In NLP, Richard Bandler and John Grinder and others developed a number of models, (and borrowed others), while most of the “modeling” done by others (including ourselves) has actually been applications of these models. We assume that most modeling tasks will also be of this nature: applications of NLP models to a specific domain and outcome, rather than creating a new model. However, if someone is able to create a new model, that will be wonderful.
   Of course there are plenty of other models available for understanding human functioning, healing, and development. Some, like crystal healing or aura balancing, do not share the epistemology of NLP–the requirement of rigorous testing, etc. Others, like standard allopathic medicine, share the NLP epistemology (at least theoretically), but their methodology and primary domain of application is quite different (though there is some overlap).
   Those who might want to explore how to adapt or describe another model to include it within NLP practice would be wise to examine the other model at all three levels (epistemology, methodology, and technology) to determine whether there is (or could be) a basic compatibility between it and NLP.

                       Your Modeling Project


     Why model something anyway? Centuries ago, people used to build barns and bridges any way they could think of. Some collapsed, others lasted until the first big wind or heavy snow, while others endured for centuries. Modeling simply enables us to do things predictably, efficiently, and effectively.
   For instance, “solution-oriented” brief therapists persistently refocus the client’s attention from problem events to “exceptions” or what we call counterexamples: times and places when the problem doesn’t exist (or at least when the problem isn’t as severe).
   However, since they don’t model the structure and process of these exceptions, they have to begin the search anew with each client–and some clients don’t seem to have exceptions, or they are very hard to find. By modeling the structure and process of exceptional behaviors or skills, NLP is gradually developing a set of “off-the-shelf” software to teach to a client. Besides being more efficient, this  process can study a resourceful exception in one person and offer it to others who (at least apparently) don’t have exceptions.
   For instance, once you know the NLP phobia cure, and how to test to be sure the person’s phobic response is a very rapid response to a set of stimuli (and therefore appropriate for the phobia cure), you don’t have to find out the exquisite detail of how the person does their phobia, something which varies enormously from person to person. Some do it by stretching time into an eternity or an endless loop, some by making the threat huge, others by making themselves very small, etc.
  The NLP model also enables us to examine other treatments for phobias to figure how (or if) they work. For example Jerilyn Ross treats phobias by asking people to relive their phobias, and as they do this she keeps them moving through the experience by asking, “And then what happened?” “And then?” “What happens next?” By doing this repeatedly, she teaches people to speed up the process and go through the phobic response very quickly. She doesn’t actually cure the phobia, but she does teach people how to get through it very quickly. After treatment you can see that the phobic response is still there, but it’s very short, so it doesn’t bother the person as much.
   Brian Weiss treats phobias using past-lives regression. Again, his clients don’t actually lose their phobias, but they become less important to them because they learn to experience them against a very long time frame of many past lives and many future lives to come. Seen against this extensive background, the phobic response becomes much less important. After treatment his clients say, “I still hate water, but it doesn’t bother me; it doesn’t matter.” This is an example of what John McWhirter has described as a “perspective” pattern. In the larger perspective of a long series of lives, the phobic response seems small and insignificant.

A. How to Start

    There isn’t “a” way to model something. A modeling process has been successful when you have a description (in injunctive language) that enables you to:

    1. Gain the skill, or transform a limitation into something more useful.

    2. Teach someone else to gain the same benefits.

    An even better test of your modeling is to teach someone else your model and see if they can teach someone else to gain the same benefits. When you can do this, you have succeeded, and how you get there is not important.

B. What to Model

    The first step is to define the skill, ability or limitation that you want to model, and the context in which it occurs. Chunking this down to a reasonable size is very important, particularly when you have limited time. Even when you have more time it is usually much more useful to chunk down to components, model each one separately, and then integrate these components into a larger model.
        One important distinction is between modeling a process that is mostly internal, such as shame or feeling bad about being criticized, in contrast to processes that are interactional, such as negotiation or teaching. Negotiation or teaching is inherently more complex, because you have at least two individual worlds and their interaction to deal with.
        It can be useful to chunk down to a particular kind of interaction, or stage in the interaction, or even to one person’s process/response in the particular interaction.
        A precise model of a small part of a process is generally much more useful than an imprecise model of a larger process–and you can build a precise model of a larger process by modeling small pieces of it and then integrating them.
   There are many possibilities for how to choose a starting point. Following are a few of the possibilities that we have found useful:

    1. Think of a particular difficulty and its resolution (for which there is not yet an NLP pattern). Usually these will be nominalizations (“difficulty,” “resolution”), and your modeling task will be to denominalize it into the processing that the person goes through, to find out “How, specifically?” the person does it. If you model a nominalized experience, it will typically be at a sufficiently general level that your model will be applicable to a wider range of people than if you model a simpler and more specific skill. However, usually as the level of generalization increases, so does the complexity of the process you will need to model.
    You can model the problem and its resolution separately–or alternately for contrast–and then model a process that will make the transition from one to the other (more on this later). This is how Connirae and I modeled the Grief, Guilt, Shame, and Forgiveness patterns.
    Remember that your model can only be as good as the experiences that you choose to model. When modeling grief, for example, we passed over people who said (often with a sigh, and shallow breathing) that they now felt “OK” about the lost person. Instead we chose people who felt (and behaved) joyously when thinking of the lost person. If we had modeled the former, we would have modeled a less-than-optimal solution. However, for practice in learning how to model, modeling a less-than-optimal example can be just as useful.

     2. Think of a particular skill that you, or your clients, want or need. Find a particularly good example of someone who has that skill behaviorally, and model what they do differently from when the same person is not able to exhibit the skill, or differently from someone who does not have this skill. This is how we modeled how to respond resourcefully to criticism.
    In selecting a model, be very cautious about people’s self-reports. For example, some people say that they are good at motivating themselves because they are so aware of the hour-long process they use to get out of bed! Others will say they are not good at motivating themselves because they can’t continue to motivate themselves at the end of a highly active and productive 18-hour day! Find someone who actually exhibits, or can demonstrate to you, the skill or quality that you want to model.

    3. Explore the structure of anything that you are curious about or fascinated by. This is how Connirae and I modeled how people represent time and criteria, and how I modeled the structure of self-concept. This is potentially much more generative, but it may also be more complex, and the applications, uses, and benefits are usually not clear in advance.

    4. Look and listen around you for someone who is noticeably good at something or consistently exhibits a pleasant or useful attitude, and model that. This may be a particularly useful option. Although consistent attitudes typically generalize widely, they can be fairly simple in structure/process. Ror example, a ukseful attitude may be a consequence of a single fundamental presupposition. There are plenty of attitudes the world could use more of (gratitude, appreciation, tenacity, friendliness, tolerance, love, respect, connection, equality) and plenty of attitudes the world could use less of (scorn, hatred, meanness, superiority, inferiority, coercion/manipulation, imposition, distance, grouchiness, etc.). You can think of people in your life whose attitude you particularly like or dislike, and model that. I got interested in modeling self-concept by my dislike of pompous people whose self-esteem seemed to be much too high!

     5. Notice the universal form of an individual solution: when a client presents you with a difficulty and you find a solution that works for them, chunk up to a more generalized form, and try applying the solution to others. This is how Connirae modeled a number of processes: Self healing, Core Transformation, Parental Timeline Reimprinting, Timeline Recoding, and Naturally Slender Eating.

     6. Model a useful change that someone made spontaneously. Find out the characteristics of before and after, and how the transition was made. I have rediscovered the Swish Pattern, Content Reframing and Change History a number of times doing this. Although it did not result in a new pattern, it was a wonderful way to gain experience and flex my modeling muscles.

     7. Model a skill of your own that other people have commented on, but you don’t know clearly how you do. Ask someone who doesn’t have this skill, and wants it, to gather information about it as their project. Since it is so natural to you, there will be many aspects that will be totally unconscious and presupposed. Someone else asking questions from a perspective of not being able to do it will be likely make them obvious.

C. How to Proceed

                       1. Contrast

     Some kind of contrast will be extremely useful in helping you find the crucial distinctions operating. Whenever possible make everything the same except the presence or absence of what you are modeling.
a. You can compare the same person before and after they made a change–whether spontaneous or deliberate.

b. You can compare two recent experiences in the same person when they did, and didn’t, have the skill or quality you are modeling.

c. You can compare two people, one of whom has it and the other doesn’t.

                      2. Selecting a counterexample.

     If you are modeling a problem state, for example, you don’t want to select any counterexample. You need a counterexample that has all the features described for the problem state except that the person’s response is useful and life-affirming. This will be an immense help in disregarding all the elements in the two experiences that are the same, and are irrelevant to success/failure. However, later you may need to go back and identify other supporting elements that are necessary, but not sufficient, and since they were present in both experiences you disregarded them.

                       3. Characterizing the experience and its counterexample. It is usually helpful to start with the essential large-chunk features of the states or skills you are modeling. What are the most obvious differences between the two? Is one mostly internal (catatonia) and the other mostly external? What are the cues or triggers that send the person into one or the other? What, to paraphrase Bateson, are “the differences that make a difference?” What overall strategy sequence does the person go through? Then chunk down to the smaller steps, and characterize them using any and all NLP distinctions and methodologies that you have available to you. Among the ones that are usually very useful are:

                        Rep. Systems

                        Strategies (sequence)

                        T.O.T.E. (Test, Operate, Test, Exit), or

                        G.E.O. (Goal, Evidence, Operation)



                        Meta-program Sorting principles



                        Perceptual Position and Alignment

                        Attentional Shifts: self/other content/context




           4. Content:

    Most of the distinctions above are pure process differences and do not contain specific content. However, most real-world skills also requires knowledge of content. A geologist needs to know about rocks, chemistry, physics, etc., and a negotiator may need to know about corporate structure, contracts, interest rates, time to develop a product, etc. This content knowledge is essential for the good judgement required in carrying out the process distinctions in your modeling. These are often overlooked in the focus on process, and need to be included as a part of your modeling. For instance, an editor needs to know the letters of the alphabet, and how to read and speak the language involved. Even if it seems totally obvious to you, include required content areas in your modeling description.

                       5. Designing a Transition

    When you have characterized the differences between the problem state and the desired state, or between having or not having a useful skill, this will usually suggest what changes are required to get from one state to the other. How can you design a sequence of changes to make the transition smooth, efficient, and effective? Keep in mind that a given set of changes may be very difficult when made in one sequence, and very easy when done in a different order. If there are a number of shifts to be made, decide which will probably be easier or more comfortable to make first, and then experiment to find out the best sequence of these shifts. Modeling someone who spontaneously went through a transition successfully will provide one effective sequence, but there is no guarantee that it is the best sequence.

    At this point you should have an outline of a model of how to achieve the desired outcome. It is probably missing some distinctions and there will be certain clients for whom it won’t work, but it will work in at least some cases.

                       6. Testing and Refining Your Model. 

Some refining can be done conceptually, but trying out the model with yourself and others is the best way to learn how it can be improved. By trying out your model with additional clients, you can discover additional useful features.

     a. Congruency. Try out your model with yourself. What problems could occur? How can you modify the process so these problems are excluded? Are all the positive functions of the problem state preserved? For example, if someone feels comfortable while public speaking by negatively hallucinating the audience, this will interfere greatly with a lively, connected presentation. An alternative way of feeling comfortable will be much more useful. Are there any supporting elements, or processes, reframes or preframes, etc., that you can add that would make this process even more positive, attractive, and beneficial for the person?

     b. Streamlining. The process you modeled from the counterexample or exceptional model may have steps or aspects that are redundant or superfluous, and may even interfere with the desired outcome. Is there anything you can leave out, yet still get the desired result? Perhaps someone repeats a question inside, or shifts posture, etc., and this only delays the response.

     c. Amplifying. How can you add to the process to make it more robust and enduring? This is best discovered by noticing exactly where the process fails with specific clients, and what you have to change to make it work. By building this into the process you can extend the range of successful applications. For instance, the phobia cure will not work well with some people because of postural anchors that prevent full dissociation. Perceptual position misalignment can also interfere. Adding these elements in, either as an earlier step in the process, or as “troubleshooting” followups can make the phobia cure work successfully with a much wider range of people. Sometimes the process can be amplified by changing the sequence of states or representations, or by changing the tempo of the sequence (or both).

                      7. Different Contrast

    At this point it can be extremely useful to compare your model of an exceptional skill with:

       a. Someone who is only moderately skilled, to gain more understanding of the relative contribution of individual components to the overall ability, and to highlight aspects that may interfere, or that were not obvious in your previous modeling.

       b. Someone else who is also exceptionally skilled, to learn different ways to do a particular component of a process, and/or to learn additional supporting elements that your first model never learned–and that you can teach them to improve their performance even more. This potential improvement can be a useful incentive to offer a highly-skilled person to interest him/her in participating in your modeling project. Another incentive is that when you are successful, they will have an explicit model that they can teach to clients or associates, to their benefit.

       c. Special cases. Some clients will need more than a small adjustment to deal with objections, concerns, problems, or unique aspects. Often you can simply add a “standard” step that checks for congruence (“ecology”) or that preframes or reframes common objections, so that the model can be successfully applied to a wider range of clients without further modification.

       Refining could theoretically go on forever. Typically when you have experience with 20 or 30 clients, you will have encountered most of the variations that exist. One way to speed up this refining process is to meet with a group of people and run them all through the process at once, with explicit directions to please report any and all concerns, hesitations, objections, or difficulties to you so that you can learn about them and build solutions into your model. (A tape recorder will help you get all this information quickly, and you can review it all later.)

      Modeling is the basis for the continuing development and progress in any field. Physics began over 250 years ago; NLP only about 25. It’s a nice beginning, but so much more must lie ahead.

References1. Wylie, Clarence R. Jr. quoted in “The Imperfections of Science” by Warren Weaver. Proceedings of the American Philosophical Society, Vol. 104, No. 5, October, 1960.

2. Vaihinger, Hans. The Philosophy of “As If” Routledge and Keegan Paul, Ltd. 1924.

3. Valery, Paul. “Moralites” Collected Works of Paul Valery. Princeton University Press, 1989.
*Reprinted from Rapport: The Magazine for Neuro-Linguistic Programming (UK) Winter No. 46, p. 7