Abstract: Near infrared spectroscopy（NIRS）, as a fast and nondestructively analytic technique, is a promising tool in on-site aging condition assessment of oil-paper insulated transformers. However, some key steps such as the assessment modeling and filed applications are still in the lab-exploring stages. This paper presented the quantitative evaluation models based on NIRS to predict the degree of polymerization（DP） of insulating paper, which is directly indicating the ageing condition. The NIRS coupled with DP measurements were performed on Kraft papers with varying ageing conditions by thermally accelerating ageing method. The basic quantitative evaluation model has been proposed by partial least squares（PLS） coupled with Savitzky-Golay（S-G） convolution method. Further, three modified evaluation models by combining the outlier sample eliminating algorithm based on joint XY distances with the competitive adaptive reweighted sampling（CARS） were introduced to optimize the models. The assessments of predictive results suggest that the quantitative evaluation model established by XY eliminating algorithm coupled with CARS could self-adaptively predict the ageing condition of oil-immersed paper, with less spectral data but achieving higher accuracy. The results of field test and assessment on five power transformers show that the established model has good performance in predicting the aging state of oil-immersed paper.