Based on particle swarm optimization (PSO) and fuzzy linear proximal support vector machine (FLPSVM), a rotating machinery fault diagnosis method is proposed in this paper. Fuzzy PSVM could solve this question that standard proximal support vector machine (PSVM) is sensitive to outliers and noises in the training set. When PSVM is applied to the problem with two classes on unbalanced dataset, it tends to fit better the class with more data points. This leads to the poor classification performance. The penalty factors are designed for the deference sampling, in order to improve on classification performance. The penalty factors of FLPSVM are optimized by the canonical particle swarm optimization to avoid dependence on initial parameters and training samples. Rotating machinery fault-classification data is used to demonstrate the designed method. Firstly, vibration signals are filtered by the filters. Secondly, the energy of the deferent frequencies spectrum peaking is utilized to identify the five typical rotating machinery faults of input parameters of fuzzy linear proximal support vector machine (FLPSVM) classifier. The experiment results demonstrate that the modeling method is correct and precise.