Modeling Modeling*

by Steve Andreas

The Model Magazine, Spring, 2006

           In a recent article, Robert Dilts introduced a statement by Carmen Bostic St. Clair and John Grinder proposing a distinction between “NLP modeling” and “Analytic Modeling.” Every field progresses by making new distinctions, so it is always interesting to examine them to find out what they can offer us, and how they fit in with existing knowledge. St. Clair and Grinder write:

NLP Modeling, in the creation of the initial models that founded the field of NLP, at present and in the future of NLP, references an appreciation of and respect for two criteria that apply to modeling in NLP:

1. The suspension of any taxonomic and/or analytic attempt (all f2 transforms as described in Whispering in the Wind, see to understand consciously the patterning of the genius or model of excellence during the acquisition stage of patterning and until the following criterion is met:

2. The modeler must demonstrate the ability to reproduce the patterning of the model in parallel contexts and in such contexts elicit roughly the same responses from client with roughly the same quality and time commitment as the original genius or model of excellence prior to beginning the challenging and rewarding activity of codification of the patterning demonstrated by the modeler.

We further note that all modeling work products failing to meet these criteria are to be classified as some other logical type of model-we suggest Analytic Modeling as a general term for such work products; employing the patterning and the distinctions available in the technology of NLP applications but failing to respect the definition of NLP modeling. (6, p.2)

           What St. Clair and Grinder call “NLP modeling” is something that every normal child does when they play “dress up,” learn language or acquire any other behavior unconsciously through observation and imitation. With the discovery of “mirror neurons,” we are learning more about the neurological basis of how we learn new behavior in this way. Because this process is not unique to NLP, and because I believe that other kinds of modeling deserve to be included in the category “modeling,” in this article I will use the term “unconscious acquisition” for this process.

            I am not sure that the term “analytic modeling” adequately describes the wide variety of different modeling processes that have been used in the field by means other than unconscious acquisition. The term “analytic” sounds very “mental,” as if there was no utilization of observation, and behavioral testing, or unconscious participation, etc. in the process. As long as it is clearly understood that this term encompasses a broad spectrum of modeling processes, I am willing to use that term until someone suggests a better one.

            Paragraph 1 above requires that acquisition is done unconsciously, prior to explicit and conscious codification. Paragraph 2 specifies the results: that the modeler can duplicate the pattern of excellence demonstrated by the expert model.

            The first criterion specifies the process of creating a model, while the second criterion specifies the outcome: the modeler can duplicate the pattern of excellence.

            The field of NLP has always been focused on outcomes, and it has always valued having as wide a range of choices as possible in reaching those outcomes. The test of a good model should be the outcome that it enables, not how the model was derived.

            The outcome of a modeling process can be evaluated irrespective of how the modeling was done. Whether someone models consciously or unconsciously, receives a model in a vision, or reads it on the side of a dill pickle, is completely irrelevant to the question of how effective a model is-even if the first two methods have a somewhat better “track record.”

            In the past, St. Clair and Grinder have only specified a more inclusive criterion based solely on results: that the modeler can use a model to transfer a pattern of excellence to someone else (not necessarily the modeler). For instance, St. Clair and Grinder write:

Models are simpler creatures; the sole criterion (at least thus far accepted) for their evaluation is,

            Does this pattern/model work-that is, is it learnable and upon learning it, does the learner display behavior similar in results and quality to the source from which it was extracted. (7, p. 55)

            As we stated previously, the principal criterion for evaluating an NLP model is whether it works-this seems roughly equivalent to two issues:

1. Is it learnable?

2. Does it lead to the learner producing results congruent with the original source of the model? (7, p. 159)

Notice that these statements, published only four years ago, state that the “sole” or “principal” criterion of the effectiveness of a model is the outcome–that a skill can be successfully modeled and transferred to someone else.

            The new requirements of unconscious acquisition, and that the modeler be able to personally duplicate the behavior modeled, were not mentioned previously. At this point we can distinguish four separate processes:

            1. Acquisition or elicitation of a behavioral skill.

            2. Demonstration of this behavioral skill in the modeler.

            3. Explicit conscious codification of a model of the behavioral skill.

            4. Use of the model to transfer the behavioral skill to someone else.

1. Acquisition/elicitation Unconscious acquisition is done implicitly and unconsciously as the first step. In analytic modeling, codification is done explicitly and consciously as the first step.

         Since all behavior and communication has both conscious and unconscious aspects, even if modeling is done “consciously,” there will be many unconscious aspects, and unconscious acquisition will also have some conscious aspects. As St. Clair and Grinder point out, “These two extremes define a continuum of possibilities.” St. Clair and Grinder propose that unconscious acquisition is superior because it does not impose the bias of the modeler on the explicit codification into a model.

         The essential difference of consequence between the process of NLP modeling and Analytic modeling is the relative contributions of the model and modeler to the final work product.

         This difference resides principally in the degree of imposition of the perceptual and analytic categories of the modeler during the modeling process–in the case of NLP modeling, the imposition is minimal; in the case of Analytic modeling, the imposition is maximal. (6, p. 3)

         Firstly, this statement seems to assume that the unconscious does not have “perceptual and analytic categories.” I think there is ample evidence from hypnosis and experimentation that the unconscious does have these, and they can be at least as biased as conscious ones. So even the “minimal imposition” of the modeler in unconscious acquisition will be substantial.

         Secondly, any relevant imposition will be evident and discernable in the outcome: the behavior of the person acquiring the model will be different from the behavior of the original expert model.

         Thirdly, any “imposition” can be either harmful or beneficial. If the modeler degrades what the expert model does (either through conscious or unconscious imposition), the person acquiring the model will not be able to produce results that are as effective as the behavior of the original expert model.

         On the other hand, if the behavior of the person acquiring the model is the same or better than the expert model, we can conclude that any imposition either was irrelevant to the outcome, or improved it. If the modeler’s bias improves the behavior modeled, then that will also be evident in the results of teaching the model to someone else. Either way, the effects of any bias will be evident in the results, and the process of eliciting the model is irrelevant, unless it can be shown by experimentation that one is superior to the other.

         St. Clair and Grinder’s description also presupposes that a single expert is adequate for generating a model. Unconscious acquisition is limited to replicating what an expert has already learned to do, and this is usually somewhat less than optimal. No matter how well an expert model can do something, their behavior nearly always includes aspects or steps that are redundant, useless, or even detrimental to the effectiveness of the behavior modeled. Different experts differ in how they accomplish the same task, and every expert will have elements or aspects that other experts do not have. Some of these will support the excellence of the behavior, while others will be detrimental or irrelevant.

         When a single expert is used, unconscious acquisition is limited to duplicating that particular expert’s way of achieving the outcome. It will include any aspects that are not useful, and not include positive aspects that some other expert model would have. When the unconscious model is later codified into an explicit and conscious model, these shortcomings in the expert model’s behavior will be replicated in the model.

         In addition, unconscious acquisition is limited to situations in which there is an expert who has already developed a skill. It can’t be used generatively to develop models for something that no one is yet able to do. Other modeling methods can be useful in opening a door to totally new possibilities, based on extrapolation from existing knowledge and principles, or some other way.

         Notice that in all the alternatives discussed above, the only criterion that is useful in evaluating the model is the outcome: the results that the model produces (as St. Clair and Grinder stated clearly only four years ago).

            This leads us to consider the second process, demonstration.

2. The demonstration of the behavioral skill in the modeler.

         The outcome of modeling is to use a model to transfer a set of behavioral skills to some other person who does not already have the skill. If a modeler can demonstrate behavior that is equivalent to (or better than) that of the original expert, that is one way to satisfy the criterion that the model works. This criterion of being able to transfer a skill to someone else is essential. However, that person need not be the modeler; it can be anyone else, and the best test of a model is that it can be used to teach a wide range of people.

         There are situations in which it is clearly inappropriate for the modeler to acquire the behavior modeled. For instance, if I, at age 70, were to model an expert skier or diver, my arthritic knees would prevent me from duplicating that behavioral skill, even if I had a perfect model of a perfect expert. Those same arthritic knees would probably also prevent me from using unconscious acquisition to develop the model, but would not prevent me from doing some other kind of modeling.

3. Explicitcodification into a model.

In analytic modeling, this phase comes first. Through observing an expert, questioning, and noting the expert’s responses, an explicit conscious model is developed and repeatedly tested, first in the expert’s experience, and then in the modeler’s experience, and later in the experience of someone who is taught the model.

         Many of the very effective submodalities methods developed by Richard Bandler were the results of modeling that did not utilize unconscious acquisition (swish, phobia cure, compulsion blowout and other threshold patterns, etc.). My wife, Connirae, and I used a different kind of modeling in the early 1980s to discover the structure of timelines, how to adjust the relative importance of criteria, internal/external reference, and how to respond resourcefully to criticism.

         More recently I have modeled the fundamental foundation of all our experience and thinking in my forthcoming book, Six Blind Elephants: understanding ourselves and others. (1) In this two volume book I model the nominalization called “generalization,” using a distinction between scope and category. Scope is the extent of what we attend to in sensory-based experience (and/or memory or forecast), which can then be categorized in a wide variety of ways, which we then respond to. Although it is a simple distinction, the ramifications are endless, and it provides a way to understand all change method–in or out of the field of NLP–using the same paradigm, a sort of “unified field theory” of change.

 In St. Clair and Grinder’s model of modeling, the behavioral skill is first assimilated unconsciously and demonstrated by the modeler. Then the modeler codifies their own behavior into a explicit and conscious model that can be used to transfer the skill to someone else.

         Since the eventual goal is an explicit and codified model, the first step of unconscious acquisition is an unnecessary extra step. Since analytic modeling omits the step of unconscious acquisition of the behavior by the modeler, it is more efficient, as long as it results in an effective model.

         Since St. Clair and Grinder assert that unconscious acquisition improves the quality of the resulting model, they need to show that this is so by comparing the effectiveness of models obtained by these different processes.

         For instance, is St. Clair and Grinder’s “new code” model (7, pp. 228-246) superior to teaching someone how to spell by simple strategy installation? Does it work better to eliminate a compulsion than Bandler’s compulsion blowout model? Is it more effective than Connirae Andreas’ aligning perceptual positions model for clarifying personal boundaries?

          Alternatively, they could use unconscious acquisition to model a new solution to several problems for which other models have already been developed, such as our models for grief or shame resolution, or for transforming anger or resentment into forgiveness. If their resulting explicit model is the same as ours, then neither process of modeling could claim superiority. If the models are different, they could be tested in clients to find out which produces the best results.

         When a modeler models their own behavior, they will be applying the same “perceptual and analytic categories of the modeler during the modeling process” that unconscious acquisition purports to avoid. The modeler’s biases are not avoided, only postponed.

         In addition, most people are notoriously unable to report accurately on their own inner processes–whether skill or problem–because they are so familiar with them, making self-modeling even more subject to the modeler’s biases. As an old saying goes, “If you have flies in your eyes, you can’t see the flies in your eyes.” Usually it is much easier for someone else to model their behavior by asking questions and observing, which is one way to do analytic modeling.

         Any explicit conscious model can be examined to identify any harmful or useless aspects in order to modify them. This process is likely to be more conscious than unconscious, but ideally will include an ample measure of both, as St. Clair and Grinder describe. (7, p. 180) When an original expert is taught how to omit or improve any harmful aspects or steps, their behavior becomes even better than it was originally. I have often used this benefit as an inducement to an expert to agree to modeling of their behavior.

         If St. Clair and Grinder propose that this kind of modification can occur unconsciously, that indicates that unconscious biases are also significant in creating the resulting model, and at least some of those biases are likely to be detrimental.

         In my modeling of self-concept (3), I used many different “experts,” because I found that each one had aspects of self-concept that worked very well, while other aspects did not. If I had modeled any one of these people–either consciously or unconsciously–the resulting model would have been seriously flawed. By combining the superior aspects of many different people, I was able to create a model that was far better than any one of the people who contributed to the model.

4. The transfer of the behavioral skill to another person.

If a pattern can be transferred to another person who is not the modeler, that is an even better indication of a successful model than whether the modeler can demonstrate the skill, since another person would not have the same “perceptual and analytic categories of the modeler.” The transfer of a behavior to any other person should be adequate evidence of successful modeling of that behavior. An even more stringent test is that someone else who has acquired the model can use the model to teach the same behavioral skill to a third person.

         When a new model is offered to people who do not yet have the behavioral skill, often the model needs to be adjusted so that it fits smoothly into the rest of who they are. These adjustments can then be incorporated explicitly into the model itself, as additional optional steps to deal with these contingencies. I have described this process of repeated testing and modification in more detail elsewhere. (4)

         In conclusion, the proposed distinction between unconscious acquisition and analytic modeling is an interesting one for those who want to have more choice in how they model, but any implication that one is better than the other is specious. Good modeling utilizes both conscious and unconscious processes. Sometimes unconscious acquisition may produce a better model; at other times analytic modeling will. The evidence for this will be in the results, not the process. The only test of a model is that it works, as St. Clair and Grinder themselves stated clearly only four years ago. Now lets compare the criteria for modeling provided by St. Clair and Grinder to their statements about the major NLP models.

The Original NLP Models

         Bostic and Grinder present “NLP modeling” as the kind of modeling that began the field. “NLP Modeling, in the creation of the initial models that founded the field of NLP.” (6, p. 2) In their book, Whispering in the Wind (7), which is largely devoted to modeling, St. Clair and Grinder repeatedly bemoan the lack of modeling in the field, and belittle most of what has been described by others as modeling.

       “In particular we refer to the lack of modeling, the very activity that defines the core of this discipline NLP.” (7, p. vii)

       “The vast majority of the actual activity at present in what is loosely referred to as the field of NLP is application and training.” (7, p. 55.)

       “It is regrettable that creating variations on such themes seem to be the principle focus of much activity in NLP as opposed to modeling of new patterns itself.” (7, p. 225)

         There is considerable discussion about the difference between NLP modeling, application, design, variations, and training, and the difference between a new model and an application of an old model is given particular attention. (7, pp. 50-56) However, no criteria are presented for clearly and unambiguously distinguishing between the different terms listed above.

         The meta model is described as “the first model in NLP” (7, pp. 142-163), so it presumably satisfies their criteria for a new model. However, they repeatedly describe it as an application and adaptation of a model already existing in transformational grammar:

         The meta model can, for example, be usefully understood to be an application of the modeling of linguistic patterning inspired by Transformational Grammar (7, p. 51).

         There already existed an explicit code for capturing verbal patterning: the descriptive and formal vocabulary for syntactic studies used by professional linguists (7, p. 146)

Thirty years ago, Bandler and Grinder wrote:

         Fortunately, an explicit model of the structure of language has been developed independent of the context of psychology and therapy by transformational grammarians. Adapted for use in therapy, it offers us an explicit Meta-model for the enrichment and expansion of our therapeutic skills . . . (5, p. 19)

         Notice that the words “explicit code,” “explicit model,” and “explicit Meta-model” in these quotes all indicate a conscious and codified model, in contrast to an implicit unconscious model.

         Important advances in knowledge are often made by applying a model that has been developed in one field to another field, and I think that the application of the meta model to the context of personal change work was an excellent and very important example of this. However, the statements quoted above make it clear that the meta-model was not an original model by St. Clair and Grinder’s own definitions, and that it was not developed by unconscious acquisition. In short, the meta model does not satisfy their own definition of modeling.

         The Milton Model is described as “NLP’s third model” (7, pp. 173-183) and also as the “inverse of the meta model.”

         When we examine the code required to explicate Erickson’s verbal patterning, we find precisely the same distinctions that occurred in the meta model. . . . Thus the meta model and the Milton model are often presented as inverses. (7, p. 261)

         If the distinctions are “precisely the same distinctions that occurred in the meta model,” the Milton model is also an application of an existing model, rather than a new model. Since this model already existed, it could not have been elicited by unconscious acquisition, despite St. Clair and Grinder’s claim to the contrary, and again does not satisfy St. Clair and Grinder’s own criteria for modeling. (7, p. 180)

         Representational Systems is described as “NLP’s second model.” (7, p.164) It was discovered by noticing that the predicates (verbs, adverbs, and adjectives) that people used to describe their experience could be categorized by sensory system, and realizing that they were literal descriptions of the speaker’s experience. Although that was a very important foundation for many of the distinctions and processes in the field, that too, is an application of an existing linguistic model. The description of the discovery of representational systems (7, pp.164-172) was again not an example of modeling according to St. Clair and Grinder’s own description.

         To summarize, none of the first three fundamental models in NLP described by St. Clair and Grinder satisfy their own criteria for “NLP modeling,” a major contradiction and incongruence.


         St. Clair and Grinder’s book Whispering in the Wind states, “Our intention is to provoke a professional high quality public dialogue among the practitioners of NLP, as an integral part of these developments,” (7, p. 348) In my review of their book (2) I pointed out the some of the incongruences and contradictions above, as well as many others. I sent my review to St. Clair and Grinder months in advance of publication to give them an opportunity to respond to it. When my review was published they promised in writing to respond to it in a later issue. It is now three years since my review was published, but they have not yet responded. Their failure to participate in a dialogue contradicts their stated intention to provoke a “professional, high quality public dialogue among the practitioners of NLP,” yet another incongruence.

         St. Clair and Grinder have harsh words for teachers who are incongruent:

         1. There were a number of extremely well-trained practitioners of NLP who were themselves clearly capable of miracles (relative to the capabilities of other systems of change work) with clients; however it apparently had never occurred to them (or perhaps they simply had chosen not) to apply the patterning to themselves–that is, self-application of the patterning. Thus, my perception was that many of them were incongruent in significant contexts in their lives–there were portions of their personal and professional lives that showed absolutely no presence of the choices they busily assisted others in creating in their lives. I was not happy with this situation. (7, p. 231)

         CAVEAT: Messengers incongruent with the message they purport to bear are not listened to, nor should they be! (7, p. 366)

            Given the incongruences noted above, these statements by St. Clair and Grinder refer to themselves, and are paradoxical as well as incongruent. I invite them to clarify these contradictions.


1. Andreas, Steve. Six Blind Elephants: understanding ourselves and others, Vols. I & IIReal People Press, April, 2006

2. Andreas, Steve. Book Review: Whispering in the Wind, by Carmen Bostic St. Clair and John Grinder. Anchor Point, Vol. 17, No. 3, p. 3 March 2003

3. Andreas, Steve. Transforming Your Self: becoming who you want to be. (book) Real People Press, 2002.

4. Andreas, Steve “Modeling with NLP” Rapport: The Magazine for Neuro-Linguistic Programming, Winter, No. 46, p. 7, 1998

5. Bandler, Richard; and Grinder, John. The Structure of Magic I. Palo Alto, CA, Science and Behavior Books, 1975

6. St. Clair, Carmen Bostic; and Grinder, John. “A Proposed Distinction for Neuro-Linguistic Programming (NLP)” The Model Magazine, Vol. 3, pp. 1-3, 2005. Visit the Forum to see the discussion.

7. St. Clair, Carmen Bostic; and Grinder, John. Whispering in the Wind. Scotts Valley CA, J & C Enterprises, 2001