The main point of vibration monitoring and fault diagnosis is feature extracting and classifying. The wavelet packet transform, a perfect tool for the analysis of non-stationary signals and fault features extraction, has excellent time-frequency characteristic. The ant colony optimization, a novel swarm intelligence algorithm, has been researched abroad in recent years. This paper combines the two ways and provides a new approach to intelligent fault diagnosis based on wavelet packet transform and the ant colony optimization. The new method is evaluated using experimental signals. Vibration signal of fan roller bearings was decomposed into sub-frequency bands by wavelet packet transform and then energy of each band was calculated. The best vector is choosing as the symptom parameters reflecting the feature of vibration signals measured. The states identification for machinery diagnosis is converted to clustering problem of different states. Euclidean distance is used to measure distance in clustering algorithm. And the analysis results demonstrate that the proposed method can recognize the faults types effectively.