Based on measured water contents and spectral reflectance of the bacterial spots on tomato leaves，we attempted to estimate the water contents of diseased leaves using the Gram-Schmidt transformation algorithm． The results show R694 in visible and R761，R1446，R1940，and R2490in near-infrared wavelengths were the spectra sensitive to the variations of water contents． Non-linear regression models were then developed to predict water contents using reflectance at R1940 and R2490 using Gram-Schmidt orthogonal transformation algorithm，with high R2(0． 724)，low relative error (0． 52%) and RMSE (0． 13)． This model was proved superior to the traditional linear model． The findings of this research can provide technical supports for diseases diagnosis of tomato plants under stress．