Preference dialogues display several interestin gcharacteristics that have implications on how to design human-like dialogue strategies in conversational recommender systems. Using human-human preference dialogues as an empirical base, this paper introduces a novel data manipulation language calledPCQLthat comprises explicit descriptive, comparative and superlative preference management as well as implicit preference statements such as factual information queries. The usage of the PCQL language is demonstrated by an implementation of a music conversational recommender system.