Extended Finite State Machine (EFSM) is a powerful technique for modeling and deriving test sequences from system of interests and has been widely applied to many areas. However, test data generation from EFSM model is yet a challenging task. In this paper, we explore an efficient optimization algorithm for test data generation for specified paths in EFSM models. The proposed algorithm uses Simulated Annealing (SA) strategy to generate test data for paths from EFSM models. We verify the proposed algorithm on four classical EFSM models and compare with Genetic Algorithm (GA). The experimental results show that the SA strategy is more efficient and stable than the GA in test data generation for EFSM models.