Abstract: In order to estimate nutritional level and growth status of tomato leaves in greenhouse fast and accurately, the correlation between the nutrient content and every multispectral image feature was studied. The multispectral images of the tomato leaves were acquired in the natural sunlight condition, and then the multi-scale Retinex algorithm was adopted to reduce the imaging degradation caused by the nonflatness of the leaf surface. Based on color model, vegetation indices of NDVI and RVI, 49 characteristic parameters of multispectral images were defined and calculated. Correlation analysis and systematic cluster analysis were used to eliminate multivariate collinearity of the above-mentioned self-defined parameters and finally four optimal parameters were extracted. The stepwise multiple regression was used to develop the prediction models of the SPAD value and nitrogen content of tomato leaf. The result showed that the model had higher predictive ability. The R-Square and RMSE of SPAD model were 0.8668 and 3.997, and the R-Square and RMSE of nitrogen model were 0.7284 and 0.5130, respectively.