In order to extract the impact feature from the rolling bearing fault signal in low signal-to noise ratio (SNR), which is important for bearing fault diagnosis, a signal processing method called S transform is introduced. Firstly, the definition formulas of S transform are derived from short time Fourier transform (STFT) and continuous wavelet transform (CWT), then discretized by fast Fourier transform (FFT). S transform successfully overcomes the disadvantage of fixed time-frequency resolution in STFT, as well as the lack of phase information in CWT. The simulation results show that S transform preserves good time-frequency resolution and concentration for all frequencies of the analyzed signal, making S transform effective in extracting the impact feature from the low SNR signal, with better performance than STFT and CWT. Finally, a physical fault vibration signal of the ball bearing is processed by S transform. The results verify the validity and practicability of S transform in the periodic impact feature extraction from the vibration signal, so as to guide related fault diagnosis for rolling bearings.