This paper presents a new classification for the use of definite and demonstrative descriptions, its application in a corpus analysis and the results of this analysis. The proposed classification is based on existing literature and extended to support the generation of definite and demonstrative NPs. The corpus analysis shows in particular, that subsequent mentions of a referent can perform two functions (repeating given information and/or introducing new information). The comparison between definite and demonstrative determiners leads to preliminary data for generation algorithms.