The invention discloses an early warning method for burst pollution of source water through aquatic organism behavior image extraction, which comprises the following steps of: ensuring that raw water to be monitored continuously passes through a monitoring container with an indicating organism by using a pump, continuously and automatically monitoring the behavior of an aquatic organism in the monitoring container by using a charge coupled device (CCD) camera, acquiring and storing an activity state data sequence of the aquatic organism with time by combining an image identification system and using a computer, processing to obtain related parameters such as an activity state average activity speed, an average activity height, an average interval, average turn times and the like of the aquatic organism, establishing a corresponding database, and comprehensively comparing behavior change data under the condition of different polluted water quality with behavior data under the blank condition to realize accurate and real-time early warning. The acquired data can be transmitted, stored and analyzed in real time, an alarm is automatically given by a communication device, and the method has the characteristics of quick response, high sensitivity and real-time, on-line and in-situ early warning.