Aiming at defects of the commonly used unsupervised clustering analysis method, a new clustering analysis method based on the theory of molecular structure design was proposed. The proposed method drew a lesson from a theoretical model of molecular structure design, and it took a fault sample space as a molecular system, fault samples as atoms in the molecule system, diversities among fault samples as the measurement index of molecular potential energy. Under the influence of the interaction potential among fault samples, taking that the potential energy among samples was the minimum as a criterion, positions of fault samples on a mapping plane were adjusted to get the optimal clustering results. Vibration tests for rolling bearings under different conditions were conducted. The clustering results showed that compared with the SOM clustering method, the proposed method reduces the clustering effectiveness index dB value by 49.04%. The proposed method was also applied in clustering diesel engines’ fault vibration data. The test results showed that its clustering effect is good, it can effectively separate different faults’ data regions; the feasibility and effectiveness of the proposed method are verified.