One of the search questions regularly used by visitors to this site is Does NLP work?

Which is why the title of this FAQ contains a rather obvious presupposition: That NLP does work.
Since one element of NLP is to avoid letting important presuppositions slip past unchecked, let's take a moment to answer the question: "Does NLP work?"

The answer, as you may have guessed, is "Yes, NLP techniques and methods do work."  But we need to put a qualifier in there to make it rather more accurate, thus:

"Yes, genuine NLP techniques and methods do work."

And here's why:

When Bandler and Grinder began to develop NLP it was virtually devoid of theory (at the conscious level, at least).  One could say with complete conviction that NLP techniques "worked" insofar as they were only adopted if someone had already used them successfully to achieve specific desired results.
This did not mean that every technique would work for everybody, or that all of the techniques worked all of the time.  It was understood that one of the key skills of all of the people whose techniques become the basis of NLP - Fritz Perls, Virginia Satir and Milton Erickson - was the way they excercised care over selecting the right technique (as they saw it) for a given person, under specific circumstances and at a particular time.

Prove It!

Before going into more detail on this subject I'd just like to clarify one a point which seems to confuse numerous critics of NLP - the question of whose responsibility it is to "prove" that "NLP" works.

A response from many NLPers, whenever the question of "proof" comes up, is something like: "If you think NLP needs to be proved, you go ahead and prove it."

This is often, and wrongly, interpreted as meaning that NLPers think that it is the critics' business to prove or disprove the validity of NLP.  But that really isn't the point at all.  The real message is: We don't regard "scientific proof" as being relevant to NLP, so we're not going to waste time looking for such proof.  If, however, you think that such experiments are of some value then by all means go for it.

In other words, it isn't a question of "passing the buck," but rather a matter of denying that any genuine "buck" exists.

This annoys many members of the scientific establishment who share the misconception that nothing can be "true" unless it can be empirically verified using the scientific method*.  In practice this claim demonstrates exactly why NLP techniques (which is what we're really talking about) are not readily amenable to "scientific" validation.

(* The scientific method is simply a process of validation.  The process starts with an hypothesis which is tested by performing a clearly formulated experiment.  If the originator(s) of the hypothesis can perform the experiment several times, and produced consistently confirmatory results, details are passed to other, independent experimenters who then duplicate the experiment as closely as possible.
If these independent experiments also produce consistent, successful results then the hypothesis is promoted from hypothesis to theory.)

In NLP, however, as we've already seen, it has always been accepted that none of the techniques will necessarily produce successful results, in every context, with every subject (this is true for many aspects of psychology in general, of course).  On the contrary, there is a built in assumption that it is always necessary to deal with each situation as unique, and that it may often be necessary to pursue a course of trial and error experimentation in order to find which technique is most appropriate to deal with a particular subject in a particular context.

On this basis, when tests on a particular technique do not produce favourable results, the reality is that this does not invalidate that technique, it simply shows that the technique is not appropriate in that particular situation - and the pointlessness of trying to apply scientific methods to non-scientific subjects.
(Note:   A more detailed explanation of why allegedly scientific experiments often fail to support even the most reasonable claims made for various NLP techniques can be found in Part 9.)

The key issue to remember here is this: Not being able to validate a given NLP technique does not prove that it is false.  As all genuine scientists know, even in the most rigorous, and appropriate scientific experiments the results do not prove that the hypothesis is true or false - they demonstrate a level of probability that the hypothesis being tested is true or false.
If we bear this simple rule in mind it is immediately clear that the majority of NLP critics - those who claim that NLP has been "disproved", "falsified", etc. - are not only making inaccurate claims; they are also demonstrating a profound lack of understanding as to the inherent boundaries of scientific testing.

The Importance of Modeling

As explain in FAQ 2, NLP itself is really just modeling - just capturing the elements of how a particular person carries out a specific function in such a way that it can be taught to other people.  (By implication, then, NLP is primarily designed for modeling people who have some outstanding skill or talent which other people might want to duplicate.)

So, if you wanted to help people to become really effective estate agents (realtors), you would not, from an NLP perspective, waste time sitting them round a table and getting them to imagine what qualities it might be useful to have in that job.  Rather you would either model a top estate agent and teach the model to your students, or you would teach them how to model and then put them in a position where they could "shadow" a top realtor for themselves.

That is "genuine" NLP, as distinct from techniques which someone thinks ought to work, or might work if the wind is blowing from East Nor East and there's an "X" in the month, etc.
So the answer to "how does NLP work" is simply: By giving people a more tangible understanding of how someone else does whatever it is they do.

Everything else - virtually everything that most people think of as NLP - is simply a collection of techniques and methods and applications which can be used in the modeling process.  That is to say, techniques like the eye accessing cues, and applications such as the "Fast Phobia Cure."
Although closely associated with NLP, none of these elements are NLP; because any one of them could be discarded without changing or affecting NLP (modeling) itself.
Unfortunately there are any number of well-meant but spurious definitions floating around, such as this one which I came across recently, which only serve to confuse the issue:

"NLP is a set of guiding principles, attitudes and techniques that enable you to change ... behaviour patterns as you wish."

In fact, NLP is a tool which can be used as part of the process of changing behaviour patterns.  But you might equally well use NLP modeling simply to gain a better understanding of the person being modeled, without necessarily wishing to actually change anyone at all.

Back to the Question You First Thought of

So, back to the initial question - How does NLP work?

Given that "NLP" is simply a very intensive modelling process, it "works" by providing a model of whatever elements of someone's behaviour, beliefs, values, etc. are essential factors in how they achieve certain results or goals.
In a wider sense, NLP "works" when the model is presented in such a way that other people can incorporate it into their own activities so as to achieve a similar level of performance.

IF, however, the question: "How does NLP work?" actually means: "How do the various methods and techniques work?" then because there are so many methods and techniques, the answer is bound to be rather more vague.  For example:

How does anchoring work?
Many people assume that the NLP technique known as anchoring is simply the Pavlovian "conditioning" technique with a different label.  But this is not entirely correct.
The key feature of the Pavlov technique was that it took an automatic response to event A (producing saliva when food appears), and linked it to an essentially irrelevant event B by introducing a false "cause and effect" relationship (producing saliva when a bell is rung, because, in the past, the ringing of the bell has immediately preceded the appearance of food).  In basic behaviourist terms this is simply building an elementary "stimulus - response" structure.
In the technique of anchoring, on the other hand, we are linking a natural response in one situation - A - to a different situation (or type of situation) - B - where it is also appropriate, and possibly changing the perception of B so that the new link becomes self-re-enforcing.

Without trying to cram all of the NLP-associated methods and techniques into a single pigeonhole, it is fair to say that many of them are based around creating mental confusion, or associations, or both.

Recommended reading:   Introducing NLP and NLP Workbook