To suppress the influence of noises associated with a laser ultrasonic testing process on the detection for defects and material parameters, the time-scale filtering process of laser ultrasonic signals is studied based on Empirical Mode Decomposition (EMD). As the aliasing between useful signals and noises in Intrinsic Mode Functions (IMF) will reduce the reconstruction signal-to-noise ratio, a time-windowing method with kurtosis test strategy is proposed considering the multi-mode and broad-band characteristics of laser ultrasonic signals.With proposed method, the positions and boundaries of useful signals in IMFs near the reconstruction start are estimated by computing the local kurtosis values. Then, a Turkey-Hanning window is used to preserve useful signals and suppress the noises. Thus, aliasing removal is achieved and the reconstruction signal quality is improved. Simulation and experimental data show that the method proposed has a good self-adaptability,and it recognizes and separates the signal and noise components effectively. The signal-to-noise improvement factor is over 14 dB,which improves the original method by 3 dB and enhances the performance by 20%. Besides, as the heavier the signal is polluted, the better the improving effect is, the advantage of the method is expected to be given in high noise levels.