Abstract: Estimating individual tree volume is one of the most important steps in the process of multi-scale remote sensing retrieval of aboveground biomass. However, existing studies have paid only limited attention to the differences between branch and leaf separation, particularly for their potential effects on volume estimation accuracy. In this study, the research data were taken from the terrestrial laser scanning point clouds of 190 individual trees with various species, spatial distributions, and crown shapes. Five advanced algorithms were selected to separate branches and leaves, including TLSeparation, LeWoS, GBS, random forest (RF), and 3DSegFormer. Branch points were used for the three-dimensional tree models with TreeQSM software. The volumes of all cylinders were summed for the individual tree volume in each model. The differences between the branch and leaf separation were determined to clarify their effects on the accuracy of estimation for individual tree volume. The results showed that the five algorithms obtained higher separation accuracy for broadleaf trees than for coniferous ones. The average accuracies were 90.01% for broadleaf trees and 77.77% for coniferous trees, and the average F1 scores were 78.98% and 60.27%, respectively. Among them, 3DSegFormer and LeWoS achieved higher separation accuracy. Their overall accuracies on broadleaf trees were 93.17% and 92.14%, and F1 scores were 85.15% and 81.26%, both of which were more than 10% higher than those for coniferous trees. These values also outperformed TLSeparation, RF, and GBS, indicating that 3DSegFormer and LeWoS were more robust for branch and leaf separation of individual tree point clouds with various species and crown structures. The volume estimation showed that TLSeparation, GBS, RF, and 3DSegFormer achieved higher accuracy of volume estimation for broadleaf trees than for coniferous trees, while LeWoS showed smaller differences between the two tree types. LeWoS achieved the highest volume estimation accuracy for Datasets 1 and 4, with relative errors of 0.090 and -0.103, respectively. 3DSegFormer achieved the highest accuracy for Datasets 2, 3, and 5, with relative errors of 0.089, 0.070, and -0.195, respectively. LeWoS and 3DSegFormer showed better accuracy and stability for individual tree volume estimation across different tree species and crown structures. The volume error sources showed that the over-segmentation of branch points led to volume overestimation, while under-segmentation led to volume underestimation, with the mean errors of 0.813 m³ and -0.838 m³, respectively. Two types of errors shared partially offset each other, thereby reducing the impact of branch and leaf separation on volume estimation, with an average error of 0.075 m³. In addition, the individual tree modeling was also the main source of volume estimation error, with a mean error of 0.794 m³ for the reference branch points. Overall, the individual tree volume was estimated using branch and leaf separation algorithms and TreeQSM. LeWoS and 3DSegFormer achieved higher accuracy in both branch and leaf separation and volume estimation. This finding can provide a strong reference for dynamic monitoring of forest resources.