Abstract: In order to improve herbicide efficiency, a map-based variable weed spraying system was designed. The core of this system was S3C2410 microprocessor. In the linux environment, the system could achieve different nozzles combinations and four spraying capacity according to weed distribution map. The map was created after MATLAB software processed weeds images which collected by machine vision system. Some key issues in prescription map production were discussed, such as weed identification, weed location and area calculation, meshing, weed location, the relationship between nozzle height and distance. The spraying system was tested on indoor test-bed, when the spraying system pressure was 0.1 MPa, the conveyor speed was 0.4 m/s. The results showed that over 90% of 50 weeds was sprayed. Compared with traditional spraying methods this system could save over 60% of the herbicide. The designed spraying system provided a reference for the study on real-time variable target spraying weed robot.