An on-line transient stability prediction method based on the data measured by wide-area measurement system (WAMS) is presented. This method is composed of measurement, identification, reduction and prediction. When disturbances occur, based on the dynamic response, the order of the original system can be reduced by clustering method with the WAMS data observed. The admittance matrix parameters of the reduced system can be identified with the least square algorithm. Then future operating trajectories of angle and angular frequency of generators can be predicted by time-domain simulation. The method proposed is verified on epri-36 sample system and Northeast power system with promising simulation results.