Abstract: Landsat 8−OLI image data was used as the main information source, combined with the measured data of 32 circular Pinus densata plots with a radius of 15 m in Shangri-La, and the panchromatic band and multi-spectral fusion images were re-sampled at different scales, and based on the analysis results of the spectral feature variation of the plots at different scales, the support vector machine regression(SVR) model of the P. densata forest LAI under sampling at 4 different scales was constructed to explore the impact of different sampling scales on estimation accuracy of the P. densata LAI remote sensing. The results show that when the observation scale of the sample plot increased from 15 m to 60 m, the correlation between LAI and remote sensing variables decreased with the increase of the observation scale. The coefficient of determination of the estimation model was 0.400–0.554; the root mean square error of prediction was 0.318–0.377; the prediction accuracy was 83.51%–86.10%. When the sampling size was 15 m, the estimation accuracy was the highest, and the R² and cross-validation accuracy were the highest, respectively 0.554 and 86.10%. This study can provide a favorable reference for the selection of sampling size in remote sensing estimation of forest LAI.