Use Heckman when there is some missing, and unobtainable data. For example, if women wages is a dependent variable you can only collect that for those women who work and earn wages. That is, women who do not earn wages is unobservale. See Peters and Champoux (1979) in Decision Sciences.