In order to effectively extract the fault features of vibration signal generated from the rolling element bearing and improve the accuracy of classification, a method of feature extraction and pattern recognition is proposed based on redundant second generation wavelet packet transform-local characteristic scale decomposition (RSGWPT-LCD) and extreme learning machine (ELM). The original vibration signal is first processed by Hilbert transform to obtain envelope signal. The RSGWPT with first-stage screening processes based on the energy ratio is taken as the pre-filter process unit to reduce random noises in the envelope signal, decomposes the signal into a series of narrow frequency bands and enhances the weak fault characteristic components in the different narrow frequency bands. Then, the selected feature packets are decomposed by LCD, and the second-stage screening processes are proposed to eliminate the pseudo components of intrinsic scale components (ISCs). Applying the spectrum analysis on those desired ISCs generated by the proposed method, the fault characteristics are easily extracted. Finally, singular value decomposition (SVD) is used to decompose the matrix which consists of desired ISCs to generate feature vectors. Feature vectors are input to ELM to specify the fault type. The proposed approach is evaluated by simulation and practical bearing vibration signals under different conditions. The experiment results show that the proposed approach is feasible and effective for the fault diagnosis of rolling element bearing.