Abstract: For traditional attitude fusion algorithms of UAV (unmanned air vehicle) based on the gradient descent method, there exist some shortcomings, such as difficulty in determining the step length, slow convergence, poor dynamic performance, and sensitivity to non-gravity acceleration. For the above disadvantages, an attitude fusion algorithm based on Nesterov accelerated gradient is proposed, in which the data from gyroscope and accelerometer are fused and the impact of non-gravity acceleration is restrained. Then, several groups of comparative experiments are conducted on the experimental platform based on Pixhawk open source flight control. The results show that the error of the attitude fusion algorithm based on Nesterov accelerated gradient is less than 0.05° in the static test, the error in horizontal sliding experiments is less than 0.5°, and the algorithm performs well in angle tracking without apparent lag in the rotation test and demonstrates better performance in the actual flight test. Therefore, the attitude fusion algorithm based on Nesterov accelerated gradient demonstrates an obviously faster convergence rate than the traditional gradient descent based attitude algorithms, and a better restraint performance for non-gravity acceleration than the complementary filter and the gradient descent method. The proposed algorithm can track the variations of actual UAV flight attitude effectively.