Since the very first systems development method was defined it has been discussed how and when to select which method(s). Several books and consultants have claimed to have found the philosophers stone, but never the less it seems that there is no single method that will ever work for (nearly) all development situations. The question then arises: When to use what? And how can one help a concrete organization decide on systems development method(s)? To answer this question we undertook an action research study in ScandiBank that employs 1700 IS developers. The action research took place in three learning cycles. The first cycle started in 2001, the latter ended in 2005. We report from the three learning cycles emphasizing and explaining the learning that took place in each cycle. The result – our answer to the research question on when to use what? – is a framework focusing on the final product, with a few well-chosen trails through the “maze” of possibilities, and some rules for election of method parts.