Abstract: To solve the problem that the early weak fault diagnosis effect based on feature mode decomposition （FMD） is susceptible to the filter length L， frequency band segment K and mode decomposition number n， a diagnostic method is proposed in which a genetic algorithm is used to optimize the preset parameters of FMD， and the kurtosis， envelope entropy and modified adaptive envelope spectrum characteristic energy ratio as the comprehensive objective function. The method uses genetic algorithm to compare the comprehensive objective function values of each component signal decomposed by FMD under different preset parameters， and selects L， K and n corresponding to the maximum value as the preset parameters of FMD. The bearing fault type is determined by the envelope spectrum characteristics of the signal processed by FMD. The open bearing fault data of Western Reserve University and University of Cincinnati show that this method has good anti-noise ability and effective early fault diagnosis ability.