The CINRAD SA Doppler weather radar has the ability of detecting forest fires, however, the spatial scales of fire echoes are far less than rainfall’s, therefore it is hard to distinguish fire echoes from others manually. In order to make the fire detection and recognition possible by radar, a new method of fire clutter filtering was proposed and discussed in detail which reserves the fire echoes while removes the clutter echoes efficiently. Meanwhile, another new method of precipitation echo filtering was developed， which used the numbers of non zero velocity gates and reflectivity gates. Based on that, an automatic fire detection system was developed, which can recognize and identify fire echoes out of clutter and precipitation echoes automatically. During one and a half year’s real time test, the system detected nearly 100 of fire events in region of Wenzhou City, most of them confirmed by forest fire management departments. The accuracy of fire detection in terms of the probability of detection (POD), the false alarm ratio (FAR) and the critical success index (CSI) statistics is 79%, 4% and 76% respectively. Within the valid radius approximately 110 km away from radar station, the system can detect, position and alarm forest (wildland) fires automatically, as well as providing fire monitoring every 6 min continuously. It could be a brand new approach of forest fire detection, which can provide timely and accurate fire information to decision makers of forest fire management.