Abstract: Based on field data of 772 permanent plots in Baihe Forestry Bureau, Changbai Mountains area of northeastern China, we established global models, including Logistic and Poisson, using least square method, and local models (GWR, geographically weighted regression), including GWLR (geographically weighted logistic regression) and GWPR (geographically weighted Poisson regression), to predict distribution of natural Korean pines (Pinus koraiensis). The results showed that slope and average DBH (diameter at breast height) of trees in sub-compartment had significant influence on the distribution of the natural Korean pines, which were mainly found in the eastern and southwestern area of the bureau, and few in the north. A comparison of AICs and spatial autocorrelation of global and local model residuals showed that GWR had obviously smaller AICs and more desirable model residuals (significant decrease of spatial autocorrelation) than global models. Thus, GWR could efficiently solve spatial heterogeneity of plots and improve the accuracy of predicting occurrence probability and count of natural Korean pines. This study would provide theoretical basis for predicting distribution of natural Korean pines in large scale forest management.