The accuracy of sensor data is a critical index to evaluate the performance of Wireless Sensor Network (WSN). Outlier detection is a crucial but challenging issue for WSN. In this paper, an outlier detection approach based on variable-width histogram was proposed. The dynamic sensor data were aggregated into variable-width histograms, which avoided unnecessary data transmissions while detecting outliers. The theoretical analysis and evaluation on real WSN dataset show that this approach has high detection accuracy, and the cost is effectively reduced.