Essentially, all models are wrong, but some are useful.
From George Box and Norman Draper's Empirical Model-Building and Response Surfaces pp. (p. 424). I came across the quote during a discussion of the merits and weaknesses of personas on the AnthroDesign mailing list.
Yes, it's obvious, but I constantly need reminding; I fall all too easy into the trap of confusing the models I've created to understand a problem for the problem itself.