The nature of the relationship between sentence form and meaning has been an important concern in generative grammar from the inception of the program. Chomsky (1955) raised the question of whether transformations ‘preserve meaning’. The suggested answer was negative at that time, and the locus of interpretation was the T-marker, the entire derivational history. In the ‘standard theory’ of Chomsky (1965), it was proposed, based on work of Katz, Fodor, and Postal, that Deep Structure, a level newly proposed in that work, is the locus of semantic interpretation, though it was acknowledged that quantifiers raise certain difficulties. Those difficulties, along with similar ones involving anaphoric relations, led to the ‘Extended Standard Theory’, where Deep and Surface Structure jointly input interpretation, and soon, with the advent of traces, Surface Structure alone. In subsequent models within the GB framework, the derived syntactic level of LF becomes the sole locus of interpretation. Finally, in more recent ‘Minimalist’ Chomskyan work, there is argued to be no one level of LF; rather, semantic interpretation is interspersed among cyclic steps of the syntactic derivation, reminiscent of the LSLT proposal, though more restricted, and very similar to proposals of Jackendoff and Lasnik in the 1970’s. I will try to sort through the motivations for these changes, focusing especially on the problem of quantifier interpretation.