Missing data are often imputed for analysis, but a single imputation may be inaccurate when performing feature selection in mining data. Feature selection procedures applied to multiply imputed data demonstrate this phenomenon and suggest that multiple imputation is an important adjunct to knowledge discovery.