A distribution network always has such characteristics as running under the complicated circumstances， low standardization degree and existence of many operating modes. This paper analyzes the origin of various bad load data in supervisory control and data acquisition（SCADA） of a distribution network， including wrong data， failure data， line reconstruction and shift current. Based on these discussions， a bad data distinguishing and processing method， which is suitable for different sources of bad data， is proposed. This method is used in an actual distribution network evaluation system， as well as a medium-voltage distribution network data analysis system. The results show the practicability of the proposed method.