The Timing belt is one of important parts of modern cars. "Heat extreme" and "overloading" are two common faults of these belts, which cause fatigue, erosion, sudden failures, and consequently damages to the engine. The present article introduces an intelligent method to diagnose these two major faults of belts, using vibration signals. The method presented in this research was designed based on a combination of experimental analyses, data mining of time domain signals and the artificial neural network (ANN) classifier. To do so, firstly, vibration signals of the belt were measured by a laser sensor in three conditions: (1) normal condition, (2) belt overloading and (3) heat extreme. In the data mining stage, five statistical features including mean, standard deviation, root mean square (RMS), kurtosis and impulse factors were extracted from vibration signals. Extracted features were used as ANN inputs to the fault diagnosis and the classification of different defects of the belt. Finally, the ANN, with the average accuracy of 76%, could diagnose and classify relevant faults in the timing belt of an engine. Results showed that intelligent methods could be used to diagnose faults of belts and therefore, to protect engines from serious damages, caused by their failures.