Automated recognition of spectral signal is an importance part of spectral analysis, which jnvolves comparing spectral signals under the same conditions and obtaining relation of chemistry constitution according to similarity by pattern recognition method and computer technique, and is useful to determine the nature of spectral signal and validate spectral signal. This paper takes spectral signal according with Lambert-beer' law as object, introduces basic theory and method of recognition of spectral signal in brief, normalizes spectral signal to decrease difficulty, introduces wavelet analysis theory, puts forward the method of wavelet packet analysis to gain features of spectral signal, obtains formula to compute feature vector and error vector by statistics method, then makes use of binary tree' mode distinguishing step by step for quick recognition, and finally gives an example to explain.