One of the central tasks in natural language generation is the production of object-descriptions. An important question for this task is how to decide which properties to include in a description and which to leave out. Dale & Reiter (1995, Cognitive Science 18:223-263) discuss various algorithms they developed alone or in tandem to determine the content of a description. They argue that their Incremental Algorithm is the best one from a computational point of view (it has a low complexity and is very fast) as well as from a psychological point of view: experiments (see e.g., Levelt 1989, Pechmann 1989, Beun & Cremers 1999) have shown that humans appear to use the same principles when producing referring expressions. Dale & Reiter's primary aim was to investigate the computational implications of obeying the Gricean maxims. Due to this motivation, certain aspects of the generation of referring expressions remained underdeveloped in their work. In this talk, we flesh out a number of these aspects, without losing sight of the attractive properties of the original algorithm. In particular, we propose to include a formal notion of salience into the algorithm, and extend the coverage of the original algorithm to include pronouns, relational descriptions and bridging descriptions. We conclude with some remarks about the implementation and evaluation of the modified algorithm. This talk describes joint work with M. Theune.