Empirical mode decomposition(EMD) was currently the most effective treatment method in the areas of nonlinear and no-stationary signals, and has been applied in fault diagnosis. After over ten years of development, four issues still need to be resolved: termination criteria, margin effects, mode mixing, duration. This study proposes a novel method called fast and adaptive empirical mode decomposition(FAEMD). In this method, the algorithm structure and characteristics of the intrinsic mode function(IMF) were similar to the EMD method. Cubic spline was replaced by order statistics filter to fit the curve, and simple termination criteria significantly reduced consumption of machine time. Using this method, the signal can be decomposed quickly, accurately and effectively, and the abovementioned problems can be avoided. Fault diagnosis for rolling element bearings were applied to verify the effectiveness of the proposed method.