In this paper a Linear Parameters Varying (LPV) torque model for an automotive diesel engine is presented based on the nonlinear subspace identification method. Diesel engine is a nonlinear system and it is not easy to get a precise torque model for control through physical analysis. A LPV state space method is first introduced to estimate the diesel engine model for its speed control. The experiment results show that the model exhibits a very high accuracy for the engine torque estimation that can't be reached by the traditional identification methods. Some discussion about the model structure and its application are presented and a new noninear systems model structure is proposed.