Aiming at the end effect and mode mixing of empirical mode decomposition (EMD) in practical applications, an improved EMD method consisting of the following techniques is proposedafter analyzing several solutions. First, short signals are extended with genetic support vector regression, then an ensemble EMD combined with an alternative envelope for the sifting procedure is employed to process the extended signals. The simulation results of two nonlinear and one simulated fault signals, as well as comparisons with other EMD methods, verify the capability of the proposed method to alleviatethe end effect and mode mixing. Applying the proposed method and envelope analysis to fault diagnosis of the ball bearing with an inner raceway defect, the results demonstrate the superiority of the method in extracting fault characteristic information for engineering applications.