We have previously shown that using agonist affinity at recombinant receptors selectively expressed in clonal cells as the dependent variable in three-dimensional quantitative structure-activity relationship studies (3D-QSAR) presents a unique opportunity for accuracy and precision in measurement. Thus, a comparison of affinity's structural determinants for a set of compounds at two different recombinant dopamine receptors represents an attainable goal for 3D-QSAR. A molecular database of bound conformations of 16 structurally diverse agonists was established by alignment with a high-affinity template compound for the D1 receptor, 3-allyl-6-bromo-7,8-dihydroxy-1-phenyl-2,3,4, 5-tetrahydro-1H-benzazepin. A second molecular database of the bound conformations of the same compounds was established against a second template for the D2 receptor, bromocriptine. These aligned structures suggested three-point pharmacophore maps (one cationic nitrogen and two electronegative centers) for the two dopamine receptors, which differed primarily in the height of the nitrogen above the plane of the catechol ring and in the nature of the hydrogen-bonding region. The ln(1/KL) values for the low-affinity agonist binding conformation at recombinant D1 and D2 dopamine receptors stably expressed in C6 glioma cells were used as the target property for the CoMFA (comparative molecular field analysis) of the 16 aligned structures. The resulting CoMFA models yielded cross-validated R2 (q2) values (standard error of prediction) of 0. 879 (1.471, with five principal components) and 0.834 (1.652, with five principal components) for D1 and D2 affinity, respectively. The simple R2 values (standard error of the estimate) were 0.994 (0.323) and 0.999 (0.116), respectively, for D1 and D2 receptor. F values were 341 and 2465 for D1 and D2 models, respectively, with 5 and 10 df. The predictive utility of the CoMFA model was evaluated at both receptors using the dopamine agonists, apomorphine and 7-OH-DPAT. Predictions of KL were accurate at both receptors. Flexible 3D searches of several chemical databases (NCI, MDDR, CMC, ACD, and Maybridge) were done using basic pharmacophore models at each receptor to determine the similarity of hit lists between the two models. The D1 and D2 models yielded different lists of lead compounds. Several of the lead compounds closely resembled high-affinity training set compounds. Finally, homology modeling of agonist binding to the D2 receptor revealed some consistencies and inconsistencies with the CoMFA-derived D2 model and provided a possible rationale for features of the D2 CoMFA contour map. Together these results suggest that CoMFA-homology based models may provide useful insights concerning differential agonist-receptor interactions at related receptors. The results also suggest that comparisons of CoMFA models for two structurally related receptors may be a fruitful approach for differential QSAR.