China. In order to further improve the accuracy and stability of predicting forest volume by remote sensing, the study analyzed the relative relationship between remote sensing variables, topographic factor, forest canopy and forest volume. The partial least squares (PLS) regression model was generated from the significant variables and the space level distribution map of forest volume was constructed. The results indicated that for the PLS regression model, the number of the best principal components was 3, and canopy, elevation, slope, 6 single bands, normalized difference vegetation index ( NDVI), ratio vegetation index (RVI), TM7/ TM3, TM4 ⅹTM3/ TM2, brightness and wetness were identified as the predictors for predicting forest volume. The results showed that the determination coefficient (R2 ), relative error (RE) and the root mean square error (RMSE) between estimated value and measured one of forest volume were 0.524, 7.33% and 1.763 m3, respectively. The total forest volume in Three Gorges Reservoir Region was 1.12 ⅹ108 m3, while the total average prediction accuracy of PLS regression model reached 89.58%. The results indicate that PLS regression method can provide an effective way to improve the accuracy of predicting forest volume at large scale by remote sensing data.