Improving the efficiency of fertilization is an effective method of enhancing income for farmers, but it depends on measuring the soil nutrients accurately and rapidly. Near infrared reflectance spectroscopy (NIRS) is a fast method to detect the soil nutrients. In order to evaluate the feasibility of using NIRS to determine the soil N and P contents, the soil samples were collected from different LULC (land use and land cover) types in Daxing district, Beijing, and their biochemical parameters were determined by traditional chemical method. Then, the near infrared reflectance spectra of the samples were acquired, and NIRS models were built using partial least square regression (PLS) and Fourier transform technology for the total N and total P. The determination coefficients of cross validation for the total N and total P were 0.862 6 and 0.668 5 respectively. Ten samples were used to test the performance of the models. The coefficients of correlation between the chemically determined value and the NIRS predicted one were 0.969 8(N) and 0.830 7(P) respectively. The root mean standard error of prediction for N and P were 0.009 5(%) and 0.008 6(%), respectively. The ratios of RMSEP to SD(RPD) were 3.78(N) and 1.69(P). The results showed that the NIRS could be used to evaluate the soil N accurately and the soil P roughly.