In an environment that is full of competition, the limited resources lead to the conflict of interests which exists among the agents, therefore it is necessary to establish models of the opponent, and to accurately predict its behaviors in order to make strategies for our own advantage. The interactive dynamic influence diagrams are used to model the unknown opponent by keeping the candidate models of the opponent in the model node which can be updated over time. Then combining with the observed actions of the opponent, candidate models in the model space are pruned by sing‘abservationaction’sequences to finally identify the true model of the opponent. The experiment results show the effectiveness and feasibility of the proposed algorithm.