Stand average dominant height and mean height are important indicators for site index evaluation. Based on 78 plot data of Alnus cremastogyne plantation in Xiangxi, using quantification theory type I, 11 kinds of factors were analyzed whether they had significant effect on height ratio (stand mean height/stand average dominant height), they were place name, stand type, gradient, slope situation, slope orientation, soil thickness, natural disaster, tree number/hm2, elevation, age and crown density. If the factor had most significant effect on height ratio, it would be selected as a dummy variable. The place name and stand type, including absolute pure stand and relative pure stand were selected as dummy variables, because they had most significant effect on height ratio. The dependability of that place name and stand type had most significant effect on height ratio were 93.70% and 99.96%. Based on these dummy variables, these non-linear correlation models of stand average dominant height and mean height had been established, and the general non-linear model had also been established. The establishing model and test precisions had also been analyzed and compared with 5 indicators, they were determination coefficient, residuals mean, absolute residuals mean, relative residuals mean and precision. The precisions of non-linear model based on dummy variable with place name, non-linear model based on dummy variable with stand type and general non-linear model were 98.65%, 97.53% and 93.43%. The determination coefficients of them were 0.9389, 0.9266 and 0.9015, the residuals means of them were 2.16E-09, 5.53E-09 and 2.30E-07, the absolute residuals means of them were 1.0631, 1.1323 and 1.8522, the relative residuals means of them were 0.1197, 0.2038 and 0.2932. It showed that the non-linear model based on dummy variable with place name were more than the others, and the non-linear model with dummy variable were more than the model without dummy variable. These method for establishing model and research results maybe help to improve the model precision and establish a biology mode used in area coverage.