This paper presents a testing method for an agent system against its specification described with Statecharts. Here, an agent System means an implementation of an agent application, Its testing method generates test sequences from the specification and checks whether the agent system behaves in accordance with the specification by executing the test sequences. For specification-based testing, we have extended Statecharts in order to describe the behavior of an agent system according to the concept of Agent UML. Then, we propose an approach to generating test sequences from the extended Statecharts. Our approach makes effective use of partial order methods for considering representative sequences of equivalent classes divided from all possible sequences. Therefore, we can efficiently manage a large number of possible sequences caused by the agents' autonomy. As a result, we can reduce the number of test sequences and the number of executions to be tested.