Rolling-element bearings are widely used in rotary machinery systems. Accordingly, a reliable bearing fault detection technique is critically needed in industries to prevent the machinery system's performance degradation, malfunction, or even catastrophic failures. Bearing fault detection, however, still remains a very challenging task because most of the bearing fault related signatures are non-stationary. In this paper, a wavelet cross-spectrum (WCS) technique is proposed to tackle the challenge of feature extraction from these non-stationary signatures for bearing fault detection. The vibration signals are first analyzed by a wavelet transform to demodulate primary representative features; the periodic features are then enhanced by cross-correlating the resulting wavelet coefficient functions over several contributive neighboring wavelet bands. A Jarque-Bera statistic index is suggested for the bandwidth selection. The effectiveness ofthe proposed technique is examined by a series of experimental tests corresponding to different bearing conditions. Test results show that the developed WCS technique is an effective signal processing approach for not only stationary but also non-stationary feature extraction and analysis, and it can be applied effectively for bearing fault detection.