We study how to choose two out of p dichotomous variables in classification problems. Our results show that positive correlation may increase discrimination while negative correlation may decrease discrimination. These results were not predicted from the analysis of the same problems with normally distributed variables.