The waveform decomposition method of Lidar pulse signal is an important way to extract the waveform parameters, which provides significant data sources for retrieving the elevation, slope, roughness and reflectance of target. A waveform decomposition algorithm on variable component parameter random sampling method (WDVCM) was proposed to process waveforms with poor SNR and certain overlapping. The algorithm regarded the compounded Gaussian function as the optimization model, and achieved the decomposition and extraction of raw waveforms by generating randomly characteristic parameters and deleting or creating Gaussian component, based on the energy function and the standard deviation of fitting as the criterion for parameter optimization. About 4584 raw waveforms in a stripe of Geoscience Laser Altimeter System (GLAS) developed by National Aeronautics and Space Administration (NASA) were processed using the WDVCM. The result indicates that proportions of fitting waveforms originated from WDVCM and NASA with correlation coefficient over 0.95 are 99% and 97% respectively. Wherein, the ratio with the differences of correlation coefficient less than 0.05 is about 98%. The averages of standard deviation coefficient (SDC) of fitting waveforms provided by WDVCM and the NASA are 2.21 and 3.28, and about 89% of SDC of fitting waveforms processed by WDVCM is less than that from NASA. It proves that the WDVCM is more applicable for decomposing overlapping waveforms with better fitting effect.