This paper discusses the nature of restrictions imposed by derivational affixes on the bases in word formation. While some derivational affixes select the base of a particular syntactic category, there are affixes whose restrictions on the base are semantic in nature. Two sets of nominalization affixes in Japanese (<i>-sa</i> and <i>-mi</i>; <i>-kata</i> and <i>-buri</i>) display different and disjoint semantic and syntactic selectional properties, which operate independently from each other, as well as interact to block certain derived nominal forms. The way different types of base selection by these affixes work and interact with each other is best accounted for by a modular approach to word formation, as advocated by the Autolexical Syntax model (Sadock 1991).