Recognizing behaviors is an important step for learning by imitation. Learning by imitation, in the other hand, is considered as a method to acquire complex behaviors and as a way to provide seeds for further learning. In this paper the problem of recognizing behaviors in a robot soccer game is addressed. A behavior is a sequence of actions, which are situations intentionally produced by a robot. This approach is based on soccer games observation. When observing a soccer game, the observer can only recognize situations. This is because he doesn't know if the robot wanted to do it or not. Each situation can be part of a complex behavior, thus, it is important to find patterns in the set of situations for each robot. In this paper is showed that an accurate inference engine, for recognizing situations, can be made with fuzzy logic, and, behaviors patterns can be recognized with Self Organization Maps. Recognized situations are also useful to build a robot soccer game commentator.