Heavy metal pollution caused enormous risks to human health and ecosystem. In this paper, soil heavy metal concentration was determined through sampling from different land use types in the watershed of Binzhou. With DEM data and remote sensing image, the main routes of pollution from non-point sources, i.e., surface runoff and soil erosion of different land use types were estimated based on SCS-CN model and RUSLE model. The precision of the model was validated and analyzed and heavy metal copper (Cu), lead (Pb) pollution loads from non-point sources in the watershed of Yellow River were estimated. The results showed that big differences exited among the soil heavy metal concentration background values from different land use types in the study area: in industrial land, mining land and grassland, the background value of Cu and Pb was high; in woodland, the value was low. The total loads of Cu and Pb was about 651913kg/a and 268004kg/a. The variation ranges of heavy metal pollution loads from soil erosion were large as well: the values in sparse woodland and grassland were the highest; in woodland and irrigable land were the lowest. The value of heavy metal pollution loads from surface runoff also differed greatly: the value in grassland and sparse woodland was much higher than in woodland and dry land. The research could be used for further analyzing the characteristics of the transfer of heavy metals, studying the heavy metal loads from non point source, and adjusting the land use structure reasonably.