The literature on demonstratives is largely built around two dogmas: (1) demonstratives are directly-referential and (2) demonstratives are context sensitive. (1) cannot be sustained in the face of data from English and other languages, and there are no compelling reasons for accepting (2). If we give up the dogmas and treat demonstratives as free variables that are interpreted with regard to a pragmatically-determined assignment function, we can explain a host of otherwise problematic data, and do so in a theoretically attractive way.