Abstract
We argue that formal modeling should be the starting point for any serious development of computer systems. This claim poses a challenge for modeling: at first it must cope with the constraints and scale of serious developments. Only then it is a suitable starting point. We present three techniques, refinement, decomposition, and instantiation, that we consider indispensable for modeling large and complex systems. The vehicle of our presentation is Event-B, but the techniques themselves do not depend on it.