In this paper we analyze the issue of learning about other agents in a dynamic multiagent system with a continuous state-space. We test different types of learning models in our double auction market system. Our experimental results show that modeling other agents can be tricky. In general, an agent does not know what kind of strategies other agents adopt. An agent can learn about the general pattern of other agents' actions by observing the history data. But to learn the underlying model of those actions, an agent must make some assumptions about the form of the other agents' behaviors. If these assumptions are wrong, the agents may perform badly, regardless of the quality of their estimation methods.