The Hilbert envelope analysis and bispectrum analysis were used to ex tract frequency components of vibration signals of a rolling bearing, and the gr ay level cooccurrence matrix (GLCM) and its characteristic statistics from bi s pectrum spectrogram were obtained. Then the characteristic statistics were furth er treated by the principal component analysis (PCA) to get principal components as the input vectors for fault pattern recognition.In an artificial immune netw ork (AIN) system, the principal components were used as the antigens. A modified artificial immune network classification algorithm (AINCA) was introduced and u sed in the bearing fault diagnosis. Through the optimization of the fault antige ns,the memory antibody nets formed and the fault classification was realized by the positive selection principle based on their affinity. The result indicates t hat the modified AINCA has a high accuracy of classification.