A two dimensional correlation spectroscopy(2DCOS) was presented to optimize the characteristic variables for pesticide residues in rice,in order to improve the accuracy for the rapid detection of pesticide residues in rice based on surface enhanced Raman spectroscopy(SERS).Firstly,the original spectra were pretreated using standard normal variable transformation(SNV),then the two dimensional correlation spectrum and diagnosis spectrum were analyzed with chlorpyrifos methyl concentration as the disturbance.The characteristic peaks of chlorpyrifos methyl were optimized based on the two dimensional correlation spectroscopy and diagnosis spectroscopy.A support vector machine(SVM) model for analyzing chlorpyrifos methyl residues in rice was developed,and was compared with the PLS model.Results showed that 2DCOS was a wonderful way for screening out the characteristic peaks related to the chlorpyrifos methyl.The performance of SVM model based on 4 chlorpyrifos methyl characteristic peaks selected by 2DCOS was better than that of the PLS model.The correlation coefficient(Rp) in the prediction set was 0.96,the root mean square error of prediction(RMSEP) was 521,and the relative prediction deviation(RPD) was 3.66,which indicated that the developed model could be used for the actual estimation of chlorpyrifos methyl pesticide residues in rice.Results showed that 2DCOS is feasible for screening characteristic peaks related to chlorpyrifos methyl in rice by simplifying the model and improve the prediction accuracy.It provides a new idea for the rapid detection of food and agricultural products by Raman spectroscopy for quality and safety.