Abstract: In this paper, multivariate statistics analysis is used to create a numerical approach for the classification of forest site. This approach makes the well-known Bogelebnek’s site Grid Maps become quantitative. Forest sites on the southern slope of the Purple Mountain are classified into 8 major types, each corresponds a definite range of nutrient and water scores. The probability of errorous differentiation is only 9.7%, representing the high accuracy of the 8 site types classified with numerical classification method.