Abstract: Biomass of ecological forest in Jinyun county, Zhejiang province was estimated by multiple linear regression (MLR), partial least squares(PLS) regression, random forest regression and BP neutral network model based on Gaussian error function (Erf-BP), according to data from TM imagery and 117 permanent subcompartments forest management survey in 2010. There were 80 independent variables of geoscience and remote sensing. Results showed that random forest regression had better effect on R2, PRECISION and RMSE, while Erf-BP neural network on VR and BIAS. Comprehensive evaluation on precision and root mean square error indicated that random forest method was a better choice