Pharmacophore, two-dimensional (2D), and three-dimensional (3D) quantitative structure-activity relationship (QSAR) modeling techniques were used to develop and test models capable of rationalizing and predicting human UDP-glucuronosyltransferase 1A4 (UGT1A4) substrate selectivity and binding affinity (as K(m,app)). The dataset included 24 structurally diverse UGT1A4 substrates, with 18 of these comprising the training set and 6 an external prediction set. A common features pharmacophore was generated with the program Catalyst after overlapping the sites of conjugation using a novel, user-defined "glucuronidation" feature. Pharmacophore-based 3D-QSAR (r(2) = 0.88) and molecular-field-based 3D-QSAR (r(2) = 0.73) models were developed using Catalyst and self-organizing molecular field analysis (SOMFA) software, respectively. In addition, a 2D-QSAR (r(2) = 0.80, CV r(2) = 0.73) was generated using partial least-squares (PLS) regression and variable selection using an unsupervised forward selection (UFS) algorithm. Both UGT1A4 pharmacophores included two hydrophobic features and the glucuronidation site. The 2D-QSAR showed the best overall predictivity and highlighted the importance of hydrophobicity (as log P) in substrate-enzyme binding.