Estimating log P (logarithm of "1-octanol to water" partition coefficients) as a memure of lipophilicity for organic compounds is of considerable importance in drug discovery. Several methods have been developed for this purpose, each with its own drawbacks and advantages. In this article, a systematic comparison of three well-documented and fully computerized methods has been attempted for a set of nucleosides and bases. The fiist method (BLOGP) is based on overall molecular properties derived from a molecular orbital calculation to predict log P. The second method (CLOGP) uses fragmental lipophilicity constants with correction factors and treats log P as an additive-constitutive property. The third method (ALOGP) is based on an additivity scheme of atomic lipophilicity constants, with the constitutive factor governed by an elaborate list of atom types. However, none of these methods take into account conformational flexibility or intramolecular hydrogen bonding, which can cause substantial discrepancy between observations and predictions. A comparison of predictions from each of these methods indicates that the atomic contribution method (ALOGP with r = 0.842 and SD = 0.51) is better than other methods (with r = 0.395 and SD = 1.2 for BLOGP and r = 0.713 and SD = 0.93 for CLOGP) for this class of compounds. Our overall assessment IS that we do not have, as yet, a highly reliable, fully computerized log P prediction method applicable to flexible heterocycles such as nucleoside analogs. 0 1993 by John Wiley & Sons, Inc.