This paper discusses results from an experimental study of concept neighbourhoods in WordNet and Roget’s Thesaurus. The general aim of this research is to determine ways in which neighbourhood lattices can be derived in real time from a lexical database and displayed on the web. In order to be readable the lattices must not be too large, not contain overlapping concepts or labels and must be calculated within seconds. Lattices should, furthermore, not be too small and they should contain sufficient complexity to be interesting for the viewer. For these purposes the sizes of the lattices of different types of concept neighbourhoods have been calculated. Using the size information should help with the task of on-line generation of the lattices.