In order to quickly and accurately measure the soil total nitrogen content(STNC), 72 soil samples were collected from Beijing City for chemical measurements and spectral analysis. By correlation analysis of the actual measured nitrogen content with spectral data which wavelength is 350~2500nm, the most relevant characteristic wave bands were selected to build the STNC estimation models. To establish accurate and optimized predictive model of STNC, the spectral reflectance and absorbance were converted into firstorder differential and secondorder differential. The results showed that both spectral reflectance and absorbance had a low correlation with STNC, so they could not be used to build prediction model. Their correlations were improved by transforming them to the firstorder differential and the secondorder differential. In various transformation of reflectance, the secondorder differential and the secondorder differential of absorbance were the most relational with STNC. The maximum absolute values of correlation coefficient were 0.868 and 0.846. The most relevant characteristic bands were 425~527nm， 819nm， 1390~1391nm and 2200~2219nm. STNC models were built through linear regression and multivariate stepwise regression. The reciprocal logarithm secondorder differential model based on multivariate stepwise regression was the optimal model among the 10 prediction models established in this article. This conclusion proved that it is feasible to use multivariate stepwise method for predicting STNC. The R2 of the optimal model was 0.829, statistics value was 86.377 and the RMSE was 0.104. This model can be used to predict the STNC of clay loam soil in Beijing City.