An adaptive wavelet analysis approach based on recursive least square (RLS) algorithm is proposed for analyzing the complicated spectra emitted by the interaction between femto-second laser and gases. The simulation results on twenty-seven sets of spectral data show that the proposed method has high efficiency in data compression. The data of the original spectra with 3968 points can be compressed to 124 points, and the correct recognition rate of support vector machine (SVM) can be as high as 100%, when thirteen or more compressed spectra are used for training and the others for testing.