This paper presents a quantitative approach to evaluating the performance of complex-valued wavelet basis on vibration analysis of rolling bearings, using the cross-correlation measure. As a statistical tool for quantifying the relationship between two time series, cross-correlation analysis characterizes the performance of different wavelet bases in analyzing non-stationary time series, such as vibration signals. Theoretical foundation of using cross-correlation measure for evaluating wavelet basis is described, and experimental study has verified that the cross-correlation measure provides quantitative guidance to selecting wavelet basis for improved performance in defect diagnosis of rolling bearings.