Optimal wavelet filter is one of the most useful methods in fault diagnosis of rolling element bearings. Aiming at finding out an optimal couple of center frequency and bandwidth of the wavelet filter, and considering the impulsiveness and cyclostationarity of faulty signals, a novel method based on negentropy and multi-objective optimized complex Morlet wavelet filter was proposed. The parameters of the wavelet filter were optimized by the improved non-dominated sorting genetic algorithm (NSGA-II), to maximize the negentropy of both the envelope and the envelope spectrum. And then, the resonance band rich in fault information was determined by the average negentropy of the Pareto set for demodulation. The experiment results validated the effectiveness in extracting repetitive transients with complex interferences and in identifying the faults of wheel-set bearings exactly.