Complex degree of partial least squares(PLS) model is often increased due to the noise and redundant informa- tion in raw near infrared spectra. In order to simplify PLS model, net analyte preprocessing (NAP) algorithm was used to extract some useful net analyte signals from the raw spectra, then three NAP/PLS models of EGCG, ECG and EGC were constructed. The number of NAP factors and the number of PLS components were optimized by cross-validation. The spectral preprocessing result of NAP algorithm was compared with that of the classical standard normal variate(SNV). The predicting result of NAP is almost similar to that of SNV, but the number of PLS components factor by NAP is much less than that by SNV. This work demonstrates that NAP pretreatment can simplify the prediction models of catechin content in green tea.