To Lewis, but also to John Grinder: Lewis wrote: Sure there are some differences, but not enough for 3 to be at a higher logical level than 2. They are both conscious mind applications of tools of modelling, some of which (if not all) have been created consciously. Modelling the modeller in this way (your 3) is the same logical level as the modeller modelling the model (your 2). I do not agree with that statement Modeling at Level 2 deals with *real world* excellence: it can be done unconsciously (2nd position modeling) and once one can reproduce the behavior one can code it, by preference using notational conventions agreed upon as being part of NLP (in a nutshell, that's the type of modeling I think "Whispering" is referring to). But to me, a level II modeling process could also be done entirely consciously: you videotape the model displaying the excellence, and then analyze everything one distinction at a time: e.g. first eye-movements, then meta-model distinctions, then meta-programs, etc. Level III is the epistemological level, the level where the philosophers of science are working on to decide (1) what is the methodology which should be applied for doing modeling at level II and (2) what are the distinctions and coding conventions.  This is a meta-level: the discussion is about how to do modeling.  Unconscious doesn’t make much sense there (but as cognitive research has been proving the last decade, the “unconscious” is present, every thing that happens consciously is connected to what happens unconsciously as well…) In the early days, my interpretation is that John was using unconscious identification and distinctions he learned before NLP was created The initial conscious distinctions he was using were those cited in Whispering in the chapter on personal antecedents (e.g. the distinctions you'll find in his book “Introduction to transformational grammar”).  This is consistent with what I heard Judith Delozier saying: "John was writing the book on the kitchen table with his transformational grammar book next to it" - unfortunately, I cannot remember which NLP book she was referring to.  (It would be nice to have John Grinder comment on this). For me the Andreas example you cite is a combination of level I and level II processes: it might be initiated at level I: a practitioner is applying submodalities at level I but is being creative about it (not following an already code pattern by the book) and then someone else is trying to find other simple patterns in that (the modeling), or more probably, the practitioner is doing that afterwards himself / However, there is another possibility: it might be initiated at level II by designing a new sequence of combination of distinctions and then try out on level I what that would do. I'm not sure whether that would really work (I personally prefer to start at level I, discover that something I did was different from what was already codes as far as I know and then work my way up to level II to end up with a new application pattern).  I'm not sure Andreas doesn't do an unconscious uptake of his model when he is starting something at level I - the only way to know would be to do level II modeling with Steve Andreas as subject.  Anyway, it's my opinion that to understand a clients problem, figuring out how it works can be done by unconscious uptake (2nd position modeling) too.