This paper deals with a method for investigating the behavior of fish in a water tank. The motion of fish is described mathematically by nonlinear state equations, in which the main causes for the motion are expressed as the components of the external force. A water tank experiment is carried out for obtaining the observation records of fish behavior. The time series data of the position for each fish is calculated from the image data recorded by a video tape. The parameter included in the model are estimated by using the least squares algorithm. As case studies, results are shown for two experiments in which bitter ling and rainbow trout are used respectively