Misfire is a common fault mode of diesel engine, but the traditional fault diagnosis methods are difficult to acquire parameters and diagnose poorly. With respect to these problems, the misfire fault preset experiment of diesel engine was conducted to collect the exhaust noise and cylinder head vibration signal for fault diagnosis on a 3cylinder 4stroke diesel engine. In order to extract the weak signal in strong noise background, the double sampling stochastic resonance method was used to extract the fault feature frequency of diesel engine to diagnose the misfire fault. The study results show that noise energy can transfer to the weak signal feature of diesel engine by double sampling stochastic resonance system so as to realize the extracting of weak signal feature under large parameter. Accordingly, the early fault of diesel engine is detected effectively and the method is also useful for complex machinery vibration diagnosis.