Strong coupling，instability，high order and susceptible to interference are characteristics of a small -scale helicopter. It is difficult to acquire its accurate system model directly. The problems of the subspace and Prediction Error Minimization ( PEM) method such as how to choose the data channel to identify，subspace type，model structure are discussed. Expansion of main channel subspace model with the PEM algorithm is designed to build state － space model and obtain the parameters. The proposed method is applied to identify the system model based on the time-domain flight test data of the helicopter W- 100，and compared with the PEM method combined with genetic algorithm. The results show that the proposed algorithm has better performance in identification accuracy and execution efficiency.