Considering the inability of wireless sensor networks (WSNs) to meet the real-time transmission of a large number of vibration signals when applied to mechanical fault diagnosis, a mechanical fault diagnosis method based on multi-level hierarchical information fusion in WSNs is proposed. A cluster tree network structure is used to enlarge the coverage of network monitoring. The information fusion in WSNs is divided into three levels: data level fusion, feature level fusion and decision level fusion. Characteristic information of the raw vibration signals is extracted by terminal nodes for data level fusion. This information is then processed by cluster head nodes to obtain pattern recognition results for feature level fusion. The results are processed by the gateway to assess the mechanical equipment operational condition for decision level fusion. This experiment shows that this method can be applied well in mechanical fault diagnosis.