With the development of hydrological model, extraction of river drainage network has been a hot topic in hydrology research. River drainage network was extracted based on topographic maps or drainage maps by digitization in the early years, but the result was influenced by data source resolution. There are presently two kinds of methods to extract river drainage network based on DEM. One is to overlay the extracted river drainage network based on DEM on the river digitalized maps which came from drainage maps or vector layer of river, to make the extract drainage network more similar to the actual river networks. But the accuracy of river drainage network depends on the resolution of drainage maps or vector layer of river. The other one is based on "inflection point" to extract river drainage network, however, the assumption of "inflection point" exists the problem of choice of scale-free interval. To solve the above problems, the river source minimum error (RSME) method was presented based on DEM in this paper. First, the relationship between the distance error of the actual river source and the extracted river network source and the size of grid was established; second, the minimum distance error was adopted as the principle to solve the problem of the uniqueness in watershed drainage network extraction, and then the river network was determined. Taking Jinjiang River as an example and using DEM with 30m resolution as data source, the RSME method was adopted to extract Jinjiang River drainage network on the platform of ArcGIS9.2. The result showed that the distance error between the river source and the extracted river network source is the smallest one when the grid numbers are up to 5814 and the minimum river length is 42m, the corresponding fractal dimension is 1.389. Moreover, the result indicated that the proposed RSME method is reasonable to extract watershed drainage network.