There have periodic impulses in vibration signals of bearing, so a new method, so called time-wavelet energy spectrum entropy, is proposed for rolling element bearing fault diagnosis. Firstly, the impulse response wavelet is constructed to extract wavelet energy spectrum in time domain by using continuous wavelet transform, then energy spectrum entropy which represents vibration signals quantitatively change with time is calculated along the time axis, bearings with different faults have different variation complexity, and the entropy is different. To identify the fault pattern and condition of bearing, entropy of different fault signal could as input vectors of support vector machine. Practical examples showed the method can diagnose efficiently faults of rolling element bearings.