Based on soil reflectance spectral measured indoor by ASD Field Spec 4 and heavy metal contents including As, Cu, Pb, Zn, Cr, Cd, Hg measured in laboratory, the stepwise regression equations were established between spectral indices and soil heavy metal contents respectively, and prediction ability of models was assessed using coefficient of determination (R2), the ratio of standard error of prediction to standard deviation (RPD), root mean square error (RMSE). The results showed that the soil heavy metals (As, Cu, Pb, Zn, Cr, Cd) was well correlated with the soil organic, Fe, Mn, Al and Mg. The coefficients of determination (R2) of estimation models on As, Cu, Pb, Zn, Cr and Cd were all over 0.5, higher than the significance level. The ratio of standard error of prediction to standard deviation (RPD) of verification samples on Pb, Zn and Cr were all reached 1.4, which showed that the model had rough estimating capacity. The ratio of standard error of prediction to standard deviation (RPD) of verification samples on As, Cu and Cd were all below 1.4, which showed the lower precision. The coefficient of determination (R2) of estimation model of Hg was 0.28, lower than significance level, which showed the model is unreliable.