The human PEPT1 protein is an influx transporter known to influence the ADME profile of several bioactive compounds. A good example is acyclovir, which possesses high safety and selectivity but has a low bioavailability. The l-valyl ester prodrug of this compound is transported by PEPT1, resulting in a three-to four-fold increase in plasma levels. Therefore, knowledge of the Structure -Activity Relationships (SAR) of PEPT1 ligands can enable us to favourably manipulate the ADME properties of promising leads that suffer from low bioavailability. We applied three classification methods (naïve Bayesian classification, recursive partitioning and linear discriminant analysis) together with the descriptors best suited for each method to a large PEPT1 inhibitor/non-inhibitor dataset. This dataset was assembled from literature and was divided into three groups: the inhibitor group (K i 1 mM, 113 compounds), the unknown group (K i > 1 mM and < 4.48 mM, 21 compounds) and the non-inhibitor group (K i ! 4.48 mM, 69 compounds). The inhibitor and non-inhibitor groups were used to build the models. The Bayesian classification model together with fingerprint descriptors was found to produce the best results. Of a test set of 46 compounds, it correctly classified 89% (concordance). It correctly classified 96% of the inhibitors and 78% of the non-inhibitors. Although the model informs us only about structural requirements for inhibition and not necessarily for transportation, the SAR identified by the Bayesian model support previously published results. In addition, the results were obtained in a short-time span, without the need of three-dimensional configurations and with more compounds than in previous studies.