This paper proposes an observer based model predictive control for the torque tracking problem of automotive engines. The mean-value model involving the intake manifold pressure dynamics, torque production and rotational dynamics are built for characterizing the engine dynamic behaviors. Regarding the real torque is unmeasurable, an observer is designed based on Lyapunov stability analysis and the measured speed signal to online estimate the intake manifold pressure, and then the engine torque is calculated based on the estimated manifold pressure and the engine speed. The torque tracking controller is designed with the model predictive control scheme and the receding horizon optimization problem is solved by means of the numerical computation algorithm of C/GMRES. Finally, the experiment is performed on the full-scale gasoline engine test bench, and the results show the effectiveness of the proposed method.