The aim of this study is to elucidate quantitative structure-permeability relationship (QSPR) of various organic molecules through Caco-2 cells, and to ascertain the relationship between gastrointestinal (GI) absorption in humans and Caco-2 cell permeability. Caco-2 cell permeability and human GI absorption data were obtained from the literature. The maximum hydrogen bond-forming capacity corrected for intra-molecular H-bonding (Hbc) and Lien's QSAR model were used in this study. The latest CQSAR software was utilized in calculating the logarithm of partition coefficient in octanol/water (Clog P) and in deriving all regression equations. For 51 compounds, a significant correlation was obtained between Caco-2 cell permeability (log Pcaco-2) and Hbc, octanol/PBS (phosphate buffered saline, pH 7.4) distribution coefficient (log Doct), log MW and an indicator variable (I) for the charge, with a correlation coefficient of 0.797. When these compounds were divided into three subgroups, namely neutral, cationic and anionic compounds, much better correlations (r = 0.968, 0.915 and 0.931, respectively) were obtained using different combinations of various physico-chemical parameters. A plot of human GI absorption vs. Caco-2 cell permeability obtained from different laboratories reveals that Caco-2 cell permeability cannot be used to precisely predict human GI absorption for compounds with Pcaco-2 below 5 x 10(-6) cm/s, due to interlaboratory and experimental variabilities, and the lack of a simple correlation between human GI absorption and Caco-2 cell permeability. Caco-2 cell permeability may be estimated from the structures of drug molecules using the above-mentioned physicochemical parameters. In general, for compounds with Pcaco-2 above 5 x 10(-6) cm/s, human GI absorption ranges from 50 to 100%. This is generally acceptable for development into oral dosage form. For the compounds with Pcaco-2 below 5 x 10(-6) cm/s, careful interpretation of caco-2 cell permeability and use of internal standard for comparison are recommended. Otherwise, good drug candidates may be excluded due to incorrectly predicted poor absorption.