The algorithms to extract watershed boundary from DEM has been developed for many years. However, the depression and flat areas of DEM typically will generate the processing difficulties for most of the algorithms, especially on high spatial resolution DEM. Focus on the difficulty, this study intends to delineate the watershed boundary using a cellular automata method. The method deals with the problem in two particular rules: (1) to examine the flow direction using the D8 Algorithm (O'Callaghan and Mark, 1984); (2) to balance the water-level. After applying these two rules on the cellular automata operation, we can simulate a simple dynamic water flow model by steadily adding rainfall on DEM. Accordingly, each cellular automata operation will generate a flow direction and the following tracing procedure and the region growing method will be used to combine the watershed into regions. The operations will continue until the sinks of the DEM are filled with the water, and the optimal exits of the depression areas are determined. As a result, all the watershed boundaries led the water to the exit of the DEM, and the watershed boundaries are determined. The initial test results indicate that cellular automata method can be used effectively to extract watershed boundaries from DEM. Moreover, the cellular automata operation successfully solves the typical processing difficulties that used to occur in depression and flat areas of DEM. The successful operation of the cellular automata method shows that major application of such a method is in the high spatial resolution DEM.