In order to gain a deeper understanding of the mechanism of diesel engine misfire faults and effectively improve the diagnostic accuracy of misfire faults, a diesel engine misfire fault diagnosis method based on frequency domain features and neural networks is proposed based on diesel engine speed signals and cylinder pressure vibration signals. For three types of faults: complete misfire of a single cylinder, complete misfire of two cylinders, and certain degree of misfire of a single cylinder, the influence of low harmonic excitation torque vector on the speed signal was obtained through frequency domain analysis of the speed signal. A polar coordinate graph was proposed to display the fault characteristics under different misfire modes, and a three-level diagnostic network based on fully connected neural network was designed to achieve fault diagnosis. Applying this method to a certain type of diesel engine, the results show that it can accurately extract misfire fault information and effectively diagnose misfire faults.