The ambiguity and coupling between the fault characteristics of the rotating machinery in petrochemical, lead to difficulty of identifying the fault type.A NN model was proposed for fault diagnosis of petrochemical rotating machinery based on high-value small samples. According to the original fault feature fitting situation, fault features of high-value small samples were extracted. The high-value small sample fault feature was used to establish an efficient NN model. The gradient search technology was used to minimize the error between the actual output value and the expected output value of the network, to achieve the best classification effect.The results show that the training data of high-value small sample fault characteristics are highly consistent with the test data, and the accuracy rate of fault type identification is 983%. The method applied to rotating machinery in petrochemical shows that it is simple and effective. The high-value small sample features are extracted accurately, and the fault identification ability is strong. It can provide reference for fault diagnosis of rotating machinery in petrochemical units and other large equipment.