In early transformational generative grammar, it was assumed that all semantic interpretation would be done off deep structure, but with the proposals for the extended standard theory (EST) of Chomsky (1968, 1972) came the realization that certain aspects of semantic interpretation, such as focus and presuppostion and scope of quantifiers, must be done off surface structure. More recent developments suggest that EST did not go far enough. Given trace theory and certain other assumptions, it appears that all semantic interpretation can be done off surface structure, including case relations, for which deep structure is usually considered necessary. This article looks at semantic interpretation as it is considered in Jackendoff's "Semantic Interpretation in Generative Grammar" (1972) and shows how certain problems and inconsistencies of interpretation that arise in the case of adverbs, co-reference, scope of modal operators and thematic relations can be resolved if all interpretation is done off surface structure. The article also discusses how this might be achieved. (Author/CLK)