In this paper, we propose a behavior generation model for entertainment robot whose behavior seems to be living. Oneproblem of common entertainment robots is hard to generate various behaviors, because behaviors of the robots are generally madeby simple rule-based algorithms. The problem causes that the simplicity of the movement of the robot lets a user get tired soon. Inthis paper, we propose a behavior generation model based on principles of respondent conditioning and operant conditioning inbehavior analysis. The model has desire parameters and neural network for choosing a behavior. The desire parameters usually riseslowly and decrease by a specific behavior. Because the neural network learns the behavior decreasing more desire parameters, themodel can select spontaneously behavior. The result of behavior generation simulation shows the model has the feature of theirconditionings.