Large amounts of data are produced in system operation, and how to extract effective information from these data has become an important research topic in the industrial application. Dimensionality reduction is a way to refine the data. Because of the low efficiency of the existing methods, these methods can’t discover the internal structure of the data. Regarding these problems, a distributed method of dimensionality reduction based on clustering is proposed, which includes the following steps:(1) Clustering the data into some small classes according to the similarity between the data variables; (2) reducing the dimension of data in a small class after being clustered respectively; (3) merging the data after being reduced dimension; (4) classifying the data after being merged by support vector machine (SVM). The data in the simulation is the test data, and the results show that the methods proposed in this paper are better than the existing dimensionality reduction methods.