Social researchers are becoming increasingly aware of the difficulties in interpreting interaction terms in non-linear probability models such as the logit or probit. In a recent article, Breen, Karlson, and Holm offer a new approach to this issue. They show how coefficients from models using logit or probit can be expressed in terms of bi- or polyserial correlation coefficients. Because the correlation coefficient is a scale-invariant metric, they suggest that the correlation metric can be used as a meaningful alternative to logit or probit coefficients in a range of situations met in comparative social research. This article presents the derivations and describes the user-written program nlcorr that allows researchers to obtain, from a logit or probit model, the correlation coefficient between X, a predictor variable, and Y*, a latent outcome variableassumed to underlie the discrete dependent variable. The command also estimates partial correlations, applies to both binary and ordered logit or probit, provides analytically derived standard errors, and allows for formal comparisons of correlation coefficients across groups.