Abstract: A new method for regulating engine speed during automated mechanical transmission (AMT) gear shift is proposed using PID neural network (PIDNN). With the help of engine platform-test, Dongfeng Cummis EQB235-20 diesel engine's throttle experimental model was established. Matlab software was used to train PIDNN to approach the characteristics of throttle experimental model. Engine speed control experiment was conducted with PIDNN, which was compiled into ECU. The results prove that, compared with ordinary PID, PIDNN shows better response speed, robustness and convergence, which improves the adaptive ability of vehicles.