Abstract: The objective of this work was to establish the models for predicting branch diameters and lengths in the larch plantations. The traditional nonlinear model, including the trees variable nonlinear model, the nonlinear mixed models and the nonlinear mixed model containing the error structure of branch diameter and length were constructed based on the data of 843 standard branches collected from 95 sample trees, and the prediction accuracies were also evaluated, by the analyses of the changing rule of the branch diameters and lengths. The results indicated that the nonlinear mixed model with the error structure for the branch diameter and length prediction was the best, with R² values of 0.783 7 and 0.865 7, and RMSE values of 0.377 7 and 0.286 6, respectively, and the phenomenon of heteroscedasticity was also elminated. The validation by using independent samples also verified that this model was optimal. The model showed that the branch diameter and length were closely related to branch location within the living crown and diameter at breast height, crown width, ratio of height to diameter. The diameter and length of branches were smaller in the top but was bigger in the bottom of crown. The diameter and length of branches in the bottom of crown were more obvious in different trees.