It is difficult to construct a controller for a mobile robot so that it can adapt appropriately in dynamicallychanging environment. To solve this problem, extensive research has been actively performedto construct a controller by evolutionary method, but few results has come out about the behavioraln~echanism of the evolutionarily constructed controller. This paper attemps to systematically analyze themechanism of the controller constructed by evolution. As a result, we have found that evolution canproduce a controller that can solve a given problem properly, where given problem is divided into severalsub problems and adequate mechanisms emerge for each sub problem. The whole problem hasbeen solved through the complicated interactions of these mechanisms.