Abstract: An effective NIR spectral detection model of rice content was explored, and effective wavelengths for the detection of rice protein content between 1100~2500nm were found. Area normalization method was used to preprocess the spectrum. Principal components regression (PCR) was used to build the regression model, effective wavelengths were selected by Martens’ uncertainty test. We found that different kinds of rice flour can be distinguished well using principal components analysis. Varieties and quality of rice can be identified according to the sample scores along principal components. Regression model based on all wavelengths has the regression coefficient r=0.9923, RMSE=0.0747 in training set, r=0.9399, RMSE=0.2103 in cross validation set, and r=0.9364, RMSE=0.1607 in prediction set. Regression model based on effective wavelengths has the regression coefficient r=0.9899, RMSE=0.0854 in training set, r=0.9437, RMSE=0.2004 in cross validation set, and r=0.9079, RMSE=0.1796 in prediction set. It is feasible to detect rice protein content using NIR spectroscopy. It is also feasible to select effective wavelengths using Martens’ uncertainty test and to detect rice protein content using these effective wavelengths.