The state of one thing is always speculated by the analysis of its property values. But the results of the properties we measured are always with deviations, some of them originated from outer interference, some the measurement itself. Once those deviations exceed a certain extent, they would influence the results of data analysis greatly. In order to improve data quality as well as avoiding deleting one record because of one of its property value is extremely bad, this article introduces a method to realize the data cleansing work which can both decrease the obvious deviations and keep maximum original information. At last, this article applies this method on the vehicle exhaust emissions’ data.