To reduce signal noise, we proposed a kurtosis-skewness screening criteria (KSSC) of local mean decomposition (LMD). We chose the maximal kurtosis or absolute skewness values to reconstruct the signal, then analyzed the de-noising signal using the enhancedenvelope spectrum (EES) to obtain the characteristic frequencies. Finally, we used the proposed method to analyze the vibration signals of roller bearings under outer raceway fault, ball fault and inner raceway fault. The signal noise was greatly reduced using the KSSC and EES, indicating promising applications for the diagnosis of all kinds of roller bearing faults.