Raman spectroscopy combined with surface enhanced technology was adopted for analysis of phosmet pesticide. Continuous wavelet transforms (CWT) and successive projections algorithm (SPA) were used for Raman spectral preprocess and characteristic Raman shifts selection, respectively. Multi-linear regression (MLR) was used for spectral modeling. It is shown that enhanced chips can achieve enhanced Raman spectral signal for low concentration of pesticides. CWT can improve spectral resolution and smoothness, and remove translation error. Characteristic Raman shifts selection method of SPA can improve analytical precision, and simplify modeling variables of MLR. CWT-SPA-MLR model can improve correlation coefficient (r) of prediction from 0.823 to 0.903, and reduce root mean square error of prediction (RMSEP) from 1.640 to 1.122. CWT-SPA-MLR method can be used for constructing analytical models for Raman spectra and has good interpretability and repeatability.