Rolling element bearings are widely used in modern machinery and play an important role in industrial applications. Tough environments under which they work make them subject to failure. The classical strategy is to collect bearing vibration signals and denoise the signals to detect fault features by using signal processing techniques. Although the noise is reduced with this strategy, the fault features may be weakened or even destroyed as well. Different from the classical denoising techniques, stochastic resonance is able to extract weak features embedded in heavy noise by utilizing noise instead of eliminating noise. The single stochastic resonance, however, fails to extract the fault features when the signal-to-noise ratio of the bearing vibration signals is very low. To address this problem, this paper investigates the enhancement methods of stochastic resonance and develops a cascaded stochastic resonance-based weak feature extraction method for bearing fault detection. Two sets of vibration signals collected respectively from an experimental bearing and a bearing inside a train wheel pair are utilized to demonstrate the proposed method. The results show that the method is superior to the other enhancement methods in extracting weak features of bearing faults.