In this paper, based on artificial immune network, a novel approach to immune memory network-based fault diagnosis methodology for on-line fault diagnosis system is presented. The diagnosis scheme consists of the memory cell network and the antibody network. They are employed to work together for network establishment, immune identification and antibody learning. Meanwhile, the key parameters of the approach are analyzed with experiments. In order to test the proposed network, the vibration signal of rolling bearing is selected as raw inputs due to its simplicity and efficiency. The results of the experiment confirm the performance of the fault pattern recognition and the strategy of ?on-line? learning.