It is sometimes claimed that semantic theories can be assessed as correct in isolation, i.e. regardless of their connections with theories that concern other areas. The present paper offers a counter-example to this thesis that concerns predicates of taste (such as “tasty”, “funny”, “scary”), their semantics and application. It is argued that once one adopts a particular analysis of taste properties that reduces them to relational properties capturing the external stimulation and the recipient’s reaction, the most suitable kind of semantic theory should treat predicates of taste as indexical expressions of a sort. It is further shown that the indexical nature of predicates of taste can be preserved by contextualist theories and violated by relativist theories. It follows that once one opts for a certain analysis of taste properties one should adopt a certain kind of semantic theory.