A new neural network nonlinear PID controller (NNXPID) to enhance the performance of the system was proposed. It is easily accomplished by transmitting the proportion, integral, differential coefficient of PID to the neural network to train. The NNXPID was applied in the engine speed controller. The results from the experiment showed that the NNXPID controller has very good dynamic and static properties. Under the control of NNXPID, the steady speed regulation is lower than 5%, the fluctation rate of speed is lower than 0.5%, the instantaneous speed regulation is lower than 10%, and the time required to steady speed is lower than 5s.