AB/HIA is a "soft" filter for identifying compounds with poor intestinal membrane permeability. The analyzed data set included over 1000 drug-like compounds with experimental human intestinal absorption (HIA) values. A sequence of recursive partitioning analyses based on multiple physicochemical and structural descriptors led to the derivation of the rule-based algorithm (filter). The obtained rules reveal a simple physicochemical model of intestinal permeability; they also account for the specific effects caused by quarternary nitrogens and biphosphonate groups. Comparison of the observed and predicted values revealed very low percent of disagreement (15% false-positives and 3% false-negatives). The unusual absorption of compounds that deviated from the predicted values was explained in terms of active transport, efflux, chemical stability, chelating ability, and solubility. Most of these effects can be accounted for by new, substructure-specific rules that can be added into the existing filter. This can lead to the development of a reliable theoretical model for predicting human intestinal absorption. If combined with other models for predicting first pass metabolism, the updated AB/HIA filter can be very useful in predicting oral bioavailability.