One mechanism for an intelligent agent to adapt to substantial environmental changes is to change the structure of its behavior network. In earlier work, we developed a context-dependent behavior selection architecture that uses structure change as the main mechanism to generate different behavior patterns according to different behavioral contexts. This paper investigates how the structure of such a behavior network can be learned. We present a structure learning method based on generic algorithm (GA). The results show that given a particular dynamic environment, consistent and robust structure can be learned to allow an agent to behave adaptively.