Abstract: Based on 1534 knot data from 60 sample trees in a Korean pine plantation in Mengjiagang Forest Farm, Heilongjiang Province, China, mixed effect model of knot attributefactors (knot diameter, sound knot length, year of death of knot and knot angle) of Korean pine plantation was established using NLMIXED and GLIMMIX procedures of SAS software. The prediction accuracy of models was compared using evaluation statistics, such as Akaike information criterion (AIC), Bayesian information criterion (BIC), -2Log likelihood(-2LL), and likelihood ratio test (LRT). Results showed that all of the mixed effect models that considered tree effect performed better than conventional fixed-effect models. For knot diameter models, the model with random parameter combination of b₁, b₂ had the best performance. For sound knot length models, the model with random parameter combination of b₁, b₃ had the best performance. For the models of year of death of knot, the model with random variables of knot diameter was proved to be the optimal generalized linear mixed model. For the models of knot angle, the model with randomvariables of intercept, knot diameter, sound knot length was proved to be the optimal generalized linear mixed model. Mixed effect model was more effective than conventional fixed-effect model for describing knot attributes. The combination of knot attributes models and reasonable prunning schemes could improve timber quality of Korean pine which is one of the main commercial tree species in Northeast China.