Abstract: In order to improve the precision of forest volume estimation, the Chinese fir(Cunninghamia lanceolata)stands of state-owned forest farm in Jiangle County,Sanming City,Fujian Province were selected as the study object, and the high resolution images of ZY-3 and the images of Alos Palsar were selected as the remotely sensed data sources. The polarization radar parameters with high correlation and the texture parameters of the optimal window were combined for the volume inversion. Eight texture features of ZY-3 high resolution image were extracted by the gray level co-occurrence matrix under 5 kinds of window sizes including 3×3,5×5,7×7,9×9 and 11×11 pixels. Meanwhile, the backscatter coefficients in HH and HV polarization modes were derived from Alos Palsar images. Furthermore, ratio of the two backscatter coefficients above was computed. The texture features from 5 different windows were used as independent variable in the inversion of forest volume respectively by using stepwise regression analysis to find the optimal window. Then, the correlation between backscatter coefficients from different polarization modes and the forest volume was computed. The results showed that for the inversion model based on ZY-3 images, the optimal window was the size of 5×5, the value of multiple correlation coefficient reached to 0.869, with a root mean square error 23.38 m³/hm² and a total estimation accuracy of 80.32%. While for the inversion model integrating the ratio of backscatter coefficients from Palsar with the texture features of optimal window from ZY-3, the value of multiple correlation coefficient reached to 0.901, with the root mean square error 22.32 m³/hm² and the estimation accuracy of total forest volume 85.42%.The results suggests using bi-source remote sensing data can produce a higher precision of volume estimation on average.