Abstract: A Radial Basis Function (RBF) neural network learning algorithm based on immune recognition principle is proposed. In the algorithm, the input data are regarded as antigens and the compression mappings of antigens as antibodies, i.e., the hidden layer centers. This algorithm can choose the number and location of the hidden layer centers by applying the principles of recognition, memory and learning, and can determine the weights of the output layer by adopting the least square algorithm. The predicted results of the mechanical property of hot-rolled steel bars show that this algorithm has the advantages of less computation and high precision compared to the K-means algorithm.