Leaf nitrogen content (LNC) is a good indicator of the nutritional status of winter wheat, and remote sensing monitoring of nitrogen level in winter wheat growth period can not only grasp the crop nutrient and growth conditions, but also help to improve the yield and quality. In this study, field data of canopy reflectance and LNC of winter wheat of three critical growth stages were collected for different treatments during 2014/2015 and 2015/2016. The correlation between LNC of winter wheat and 16 spectral indices was compared and analyzed, and then 4 spectral indices of NDSI (R594, R506), RSI (R592, R506), mSR705 and mNDVI705 were selected. On the basis of this, linear regression (LR) model, multiple stepwise regression (MSR) model and random forest regression (RFR) model were constructed and validated with independent data sets in 2014/2015. To further compare the accuracy, stability and applicability of three inversion models, the robustness tests were conducted based on the independent data sets under three different conditions in 2015/2016. The result showed that the RFR model had the best estimation accuracy among the three models, and the value of R2 and RMSE in modeling set respectively were 0.962 and 0.276, and the value of R2 and RMSE in validation set were 0.898 and 0.401. In addition, the RFR model had a higher R2 and lower RMSE than the other two models under each condition. It indicated that the RFR model combined with multiple spectral indices and random forest algorithm had higher precision and applicability, so it can effectively and rapidly retrieve the LNC of winter wheat.