Abstract: In order to confirm that surface enhanced Raman spectroscopy (SERS) with flocculent colloid as substrate, combined with chemometric methods, can effectively detect pesticide residues in oranges, rapid and nondestructive detection of phosmet pesticide residues in navel oranges was studied with the help of confocal laser Raman spectrometer of Germany Bruker Optik GmbH. Detection limit of 4.113mg/L was concluded by cross validation method, and 7 methods of pretreatment of SERS spectra were carried out. After comparison, pretreatment method is the best with baseline correction at first, and then convolution smoothing, combined with partial least squares modeling. Correlation coefficient of prediction set is 0.904 and root mean square error of prediction set is 4.890mg/L. Correlation coefficient of correction set is 0.919 and root mean square error of correction set is 3.990mg/L. The results prove that the scientificity and feasibility of quantitative analysis of SERS. The study has certain reference in pesticide residue detection of export fruit and industrial production of domestic fruit.