提出了二次型多层前馈神经网络的卡尔曼滤波学习算法，并证明了该算法的收敛性。与文献［２，３］中的学习算法和经典的误差反向传播学习算法相比，新的学习算法具有更快的学习速度、良好的泛化能力，并且对学习率有很好的鲁棒性，不容易陷入局部极小点。仿真实验结果表明了新算法的有效性。 A Kalman filter learning algorithm for quadratic multilayer feedforward neural networks is proposed and the convergence of the algorithm is proved. Compared with the learning algorithm in literature [2,3] and classical error back propagation learning algorithm, the new learning algorithm has faster learning speed, good generalization ability and robustness to learning rate Sex, not easy to fall into the local minimum. Simulation results show the effectiveness of the new algorithm.