Atom/fragment contribution values, used to estimate the log octanol-water partition coefficient (log P) of organic compounds, have been determined for 130 simple chemical substructures by a multiple linear regression of 1120 compounds with measured log P values. An additional 1231 compounds were used to determine 235 ''correction factors'' for various substructure orientations. The log P of a compound is estimated by simply summing all atom/fragment contribution values and correction factors occurring in a chemical structure. For the 2351 compound training set, the correlation coefficient (P) for the estimated vs measured log P values is 0.98 with a standard deviation (SD) of 0.22 and an absolute mean error (ME) of 0.16 log units. This atom/fragment contribution (AFC) method was then tested on a separate validation set of 6055 measured log P values that were not used to derive the methodology and yielded an P of 0.943, an SD of 0.408, and an ME of 0.31. The method is able to predict log P within +/-0.8 log units for over 96% of the experimental dataset of 8406 compounds. Because of the simple atom/fragment methodology, ''missing fragments'' (a problem encountered in other methods) do not occur in the AFC method. Statistically, it is superior to other comprehensive estimation methods.