A pot experiment was conducted across two successive rice growing seasons. Canopy spectral reflectance and total nitrogen(TN) of rice were measured synchronously at rice jointing, heading and filling stages. Difference vegetation indices (DVI), ratio vegetation indices (RVI) and normalized difference vegetation indices (NDVI) with all combinations of two wavebands between 350 and 1330 nm were calculated and related to the TN, then TN estimation models were established based on the best relations. The performance of existing nitrogen spectral indices was also evaluated and compared with the new spectral indices established in this study. The influence of rice canopy spectral reflectance to nitrogen was greater than phosphorus, and showed a decrease at visible band and an increase at the near infrared region with the nitrogen rate increase. But the responses of rice canopy spectral reflectance to phosphorus were influenced by nitrogen rate. Phosphorus increased the reflectance in visible and near infrared region under the nitrogen-deficiency condition, while decreased the reflectance in the visible region under adequate nitrogen condition. The band combination of the best vegetation indices for TN estimation was similar and mainly located in the near infrared (780–1 330 nm) / visible (750–770 nm) region, the red edge (640–700 nm) and visible (450–500 nm) region. But the TN sensitive band combination of the no phosphorus added treatment and phosphorus added treatment were different. The TN sensitive region and the determination coefficients (R2) of the phosphorus added treatment was smaller and lower than the no phosphorus added treatment. RVI(FD719, FD740), NDVI(FD419, FD552) and DVI(FD707, FD713) were the best indicators for rice plant TN concentration at jointing, heading and filling stages, with the R2 of 0.87, 0.80, 0.87 and the RMSE for estimation of 1.98, 3.68, 3.47 respectively. The estimation models were hardly influenced by nitrogen and phosphorus interaction. The performance of mND705, PRI and NDVI705 were best among the existing hyperspectral indices at jointing, heading and filling stages, but the TN estimation accuracy was influenced by the phosphorus level, especially for jointing stage and the accuracy was better under the phosphorus-sufficient treatment than phosphorus-deficient treatment. The TN estimation accuracy of existing index was lower than the new spectral indices. The combination of TN sensitive bands and the TN estimation accuracy was influenced by the interaction of nitrogen and phosphorus, choosing appropriate spectral indices is necessary for promoting TN estimation accuracy.