Actually, in almost every case where I have been really successful it has been by dint of discarding almost all of the apparently relevant features of reality in order to create a “model” which has the two almost incompatible features:
(1) enough simplicity to be solvable, or at least understandable;
(2) enough complexity left to be interesting, in the sense that the remaining complexity actually contains some essential features which mimic the actual behavior of the real world, preferably in one of its as yet unexplained aspects.
I said dangerous, and the sense in which this is true is that one is laying a trap for the majority of one’s colleagues, who are too literal-minded to understand either the necessity or the reality of the model-building process. A really well-built model can often stand a great deal of weight if used judiciously, but it can never hold up against being taken completely literally.P.W. Anderson, "BCS" and Me, More and Different, page 38.
I would argue that this approach has largely been abandoned in theoretical chemistry due to the dominance of computational chemistry.