Like many 'expert systems,' the CLOGP program incorporates a set of empirically derived rules which are supported by accepted theoretical concepts but which cannot be derived directly from these concepts. Higher log P values are expected whenever a larger effective solute cavity is required, and lower values are seen when solutes contain localized dipoles or hydrogen bond acceptors. For simple solutes whose polar fragments are effectively isolated, the fragment values contain these elements, and simple addition of fragment values gives satisfactory estimations. However, in most structures showing interesting biological activity, polar fragments are not in 'isolation,' but are affected either by electron delocalization, by field/inductive forces or by intramolecular hydrogen bonding. As the Masterfile of measured log P values has been expanded through the years, it is clear that the current CLOGP algorithm (v.3.5) needs improvement, and the paper will attempt to demonstrate that most (if not all) of these improvements can be made within the present framework of the program. (Copyright (c) 1989. Alan R. Liss. Inc.)