Carbaryl is a broad-spectrum and efficient carbamate pesticide. In this study, we proposed a surface-enhanced Raman scattering (SERS) approach for quantitative and qualitative analysis of carbaryl residues in pakchoi. Density functional theory(DFT) calculations with Gaussian 03 using B3LYP/6-311G basis sets were executed. The experimental vibrational spectrum and theoretical spectrum of carbaryl were contrasted for its assignments of Raman peaks. Magnesium sulfate, PSA, carbopack and C18 were used to remove the influences of fluorescent substances in pakchoi. The limit of detection can achieve the standard of 0.976 mg·L-1 for carbaryl pesticide residues in pakchoi. Primordial spectra were pretreated by three methods of MSC, SNV and Normalization, and then the spectra were used to construct the pesticide residues models by the method of Partial Least Squares (PLS). Based on the results of PLS, carbaryl residues extracted from pakchoi can be predicted by the MSC model with a lower root mean square error of prediction (RMSEP=1.71 mg·L-1) and higher correlation coefficient in the prediction set (Rp=0.986 5) value. It shows that the model of MSC can accurately predict the carbaryl pesticide residues extracted from pakchoi. Five unknown carbaryl concentration pakchoi samples were prepared for prediction model precision, and the values of relative deviation were calculated to be between 1.98% and 7.28%, and the predicted recovery rates were calculated to be between 95.73% and 107.28%. The T value is 0.397, which is smaller than t0.05, 4=2.776. These demonstrate that there is not evident difference between the measured and predicted values. This study illustrates that SERS method serves as an efficient method for the detection of carbaryl pesticide residues extracted from pakchoi quickly and reliably.