Seismic horizon picking plays an important role in the seismic exploration. Most of the present picking methods can only recognize horizons as some discrete disordered points, not a connected line, and work not well under the strong noise. A new horizon picking method based on the edge detection and the Kalman filter is proposed. Firstly the Canny edge detection is applied in the seismic profile to detect the points of horizons, which can include false points and noise. Then the points of horizons, as the measurement data, are input into the Kalman filter to track and recognize the seismic horizons trace by trace. Finally the multiple horizons can be recognized automatically and simultaneously. The proposed method can intelligently link the relevant points into a trajectory, and get a series of time and space data of every horizon. The simulation on the synthetic record and the practical picking on the real profile show that the proposed method can improve the continuity of horizons, and work well under the background of strong noise.