A geostatistics method combined with Geographic Information System was applied to analyze the spatial variability of soil heavy metals and to perform source identification. One hundred and seventy four surface soil samples were collected from a contaminated site with an area of 10.9 km² and analyzed for concentration of seven kind of Cu, Zn, Pb, Cd, As, Se and Hg. The results of semivariance analysis showed that they have similar spatial distribution characteristics, and concentrations of the metals were correlated in a given spatial range. Due to the special characteristics of the area such as distribution of the point pollution sources, the spatial autocorrelations of the metals increased, which was different from that of other unpolluted or slightly polluted areas. It was believed that compared with parent materials the copper smelters were the most important sources for soil heavy metals in the area.