Local mean decomposition (LMD) is an effective timefrequency analysis method, which is widely used for fault diagnosis. However it is very sensitive to noise in decomposition. In order to overcome the interference of the noise, a new method for rolling bearing signals processing integrated by nonlocal means (NLM) and LMD was proposed. Firstly using NLM for signal denoising in preprocessing by this method, then applying LMD to denoise signals as inputs, and correlation was analyzed for PF components generated from decomposing and denoised signals to choice the PF components. Afterwards, envelope spectrum analysis was applied on useful PF components. And the method was applied on rolling bearing fault diagnosis of characteristic extraction. The results show that the method can effectively extract fault characteristic of rolling bearing, and realizes its fault diagnosis.