In the last few decades, remote sensing techniques has become efficient and economical approaches for forest area analysis, including aerial photogrammetry, satellite imagery, SAR(synthetic aperture radar) and LIDAR (Light Detection and Ranging). From the above several methods, only LIDAR can provide information of the tree structures which is very useful for forest analysis. In antithesis to conventional airborne multi-echo LIDAR, full-waveform LIDAR systems are able to record the whole emitted and backscattered signal of each laser pulse. These full waveform LIDAR signals can be considered as Gaussian mixture and each Gaussian can be thought as substructure of the tree. The Gaussian mixture problem can be solved by Gaussian decomposition, however, using Gaussian decomposition must known a prior I first,so it can get the number of Gaussian,but it ’s usually unavailable. In this study, we propound a method to estimate the number of Gaussian by Empirical Mode Decomposition and both synthetic and real data is adopted for performance analysis.