This analysis shows that multivariate generalizations to the classical Heckman (1976 and 1979) two-step estimator that account for cross-equation correlation and use the inverse Mills ratio as a correction-term are consistent only if certain restrictions apply to the true error-covariance structure. We derive an alternative class of generalizations to the classical Heckman two-step approach that conditions on the entire selection pattern rather than the selection of particular equations and, therefore, uses modified correction-terms. This class of estimators is shown to be consistent. In addition, Monte-Carlo results illustrate that these estimators display a smaller mean square prediction error.