A heading estimation algorithm based on secondorder divided difference filter (DDF2) was presented to obtain the high accuracy of the heading angle in the indoor pedestrian navigation. Through establishing a quaternion attitude movement measurement model, the heading angle calculation was realized by using the DDF2 to fusion the measured data of the gyroscope, accelerometer and magnetometer. By using the process noise and measurement noise as the design parameters, the adaptive noise covariance matrix is constructed to minimize the covariance error estimation. By processing the data obtained from the rectangular reference path, the dynamic errors from the combination of accelerometer and magnetometer as well as the gyro heading estimation algorithm were 13.6°, 6.9° respectively, while the dynamic error was 2.3° after DDF2 heading estimation algorithm was used. The experimental results showed that the proposed algorithm has effectively improved the accuracy of heading estimation, reduced the effects of gyroscope drift, the linear acceleration of the vehicle and the local magnetic disturbance on the heading estimation.