Abstract: A total of 1087 sets of data from 30 plots of Chinese fir (Cunninghamia lanceolata) plantations were collected from Mazongling forestry farm in Dabie Mountains of Anhui Province. Seven commonly used height-diameter (H-D) models (i.e. linear, Chapman-Richards, Logistic models, etc.) were selected and fitted by the least square method to obtain the optimal basic model (equation 11, a Chapman-Richards model with variable D only). Based on this optimal basic mo-del, we built up the H-D model (equation 12) with two stand variables [mean height of dominant trees (DH) and density). Meanwhile, with the consideration of plot random effect, the mixed mo-del, called equation 13 and 14, which based on equation 11 and 12 were constructed, using the power and exponent functions to eliminate heteroscedasticity. Then coefficient of determination (R²), root-mean-square error (RMSE), mean absolute error (MAE) and mean absolute percen-tage error (MAPE) were used to evaluate their abilities of model fitting and prediction for determining the best model. The results showed that the fit accuracy of the model with stand variables(equation 12) (R²=0.863, RMSE=1.381, MAPE=0.971) was better than the basic model (equation 11) (R²=0.827, RMSE=1.554, MAPE=0.101). For error variance, the power function and the exponent function could eliminate the heteroscedasticity, but the former was better than the latter. The mixed models (equation 13, 14) had better fitting and prediction precision than equation 11, 12. There was no significant difference between the two mixed models (equation 13, 14) in fitting and prediction precision. In application, due to a better description of H-D relationship between different stands, the mixed-effect model (equation 13) could be used to predict tree height for Chinese fir plantations with higher precision.