The present paper researches the automatic measurement of the physical parameters ofthe stellar spectra. It is an important problem of the automatic processing of mass spectral data in the large-scale survey plan. The basic steps of the program in this article are: at first, the stellar spectra are decomposed by multi-scale Harr wavelet. Secondly, wavelet coefficients are chosen as the feature vectors of the spectrum. Finally, Non-parameter estimation is employed for estimating physical parameters of the stellar spectra. Studies show that the original spectrumonly needs to be decomposed by four-level Harr wavelet. If the wavelet coefficient at the fourth level is chosen as the wavelet feature of the spectrum, the surface gravity and effective temperature is estimated better. If the wavelet coefficient at the first level is chosen as the wavelet feature of the spectrum, the metallic abundance is estimated better. The authors use the spectral data in the literature ELODIE library to test the effectiveness of the method. When the wavelet coefficient is chosen as the feature vector of the spectrum, the experiment results show that the proposed method is robust and features high accuracy for the automatic measurement of the surface gravity, the effective temperature and the metallic abundance.