The prerequisite of solving the problem of rough handling of express parcels in logistics process is to effectively identify behaviors of rough handling of express parcels. For this reason, the work aims to propose an intelligent recognition method of rough handling of express parcels based on acceleration distribution characteristics. The data acquisition device integrated with the three-axis acceleration sensor was used to collect and intercept the acceleration data of the express parcels in the case of potential abnormal operation in real time. And then the potential abnormal data were uploaded to the server and the distribution features were extracted on the server. Finally the feature matrix was sent to the neural network for classification to get the result of the operation category of the express parcels. Experiments showed that the multi threshold interception method proposed in this paper can effectively intercept the potential abnormal data, and the acceleration distribution used as the feature can effectively classify the rough behavior. Among them, when using BP network as pattern recognition classifier, the classification accuracy can reach 93.6%, and when using CNN as pattern recognition classifier, the classification accuracy can reach 95.3%. The rough handling recognition method proposed in this paper is accurate, fast and has excellent online real-time performance. Based on this method, a database of recognition results for rough handling of parcels can be constructed to provide data support for further improving the quantitative evaluation system of the service level of express delivery enterprises. These data can also be used for in-depth analysis of the causes of rough handling of parcels, thereby proposing targeted solutions to reduce these behaviors.