Abstract: By using the branch analysis data of 955 standard branches from 60 sampled trees in 12 sampling plots of Pinus koraiensis plantation in Mengjiagang Forest Farm in Heilongjiang Province of Northeast China, and based on the linear mixed-effect model theory and methods, the models for predicting branch variables, including primary branch diameter, length, and angle, were developed. Considering tree effect, the MIXED module of SAS software was used to fit the prediction models. The results indicated that the fitting precision of the models could be improved by choosing appropriate randomeffect parameters and variance-covariance structure. Then, the correlation structures including complex symmetry structure (CS), first-order autoregressive structure \[AR(1)\], and firstorder autoregressive and moving average structure \[ARMA(1,1)\] were added to the optimal branch size mixed-effect model. The AR(1) improved the fitting precision of branch diameter and length mixed-effect model significantly, but all the three structures didn’t improve the precision of branch angle mixedeffect model. In order to describe the heteroscedasticity during building mixed-effect model, the CF₁ and CF₂ functions were added to the branch mixed-effect-model. CF₁ function improved the fitting effect of branch angle mixed model significantly, whereas CF₂ function improved the fitting effect of branch diameter and length mixed model significantly. Model validation confirmed that the mixedeffect model could improve the precision of prediction, as compare to the traditional regression model for the branch size prediction of Pinus koraiensis plantation.