Abstract: 【Objective】 Image textural features derived from optical remote sensing images have been widely used in forestry. Because window sizes can significantly influence the correlation between forest structural diversity and image textural features, in this study we attempted to determine the optimal window size based on the correlation coefficient for accurate estimation of forest structural diversity. 【Method】We first calculated forest structural diversity indices for 68 national forest inventory plots in the Guangxi Zhuang Autonomous region. Secondly, based on the SPOT-5 panchromatic band, we extracted textural features using different window sizes(3×3, 5×5, 7×7, 9×9, 11×11, 13×13 and 15×15). Finally, we performed Pearson correlation analysis between image textural features and forest structural diversity indices to determine the optimal window size. 【Result】The results showed that the correlation between forest structural diversity indices and textural features, i.e. entropy, homogeneity, angular second matrixes, contrast, dissimilarity, variation and mean value, exhibited fewer significant differences in response to changes in window size. In contrast, the change rate of the correlation coefficient between DBH standard deviation and image correlation index was as high as 86%. 【Conclusion】 Consequently, we determined the optimal window size to be 11 × 11 pixels.