A series of cyclic guanine derivatives, phosphodiesterase type 5 (PDE-5) inhibitors, have been modelled using an image-based approach for quantitative structure-activity relationships (MIA-QSAR). The calibration model showed to be robust with a R(2) of 0.864 using five PLS components. The predictive ability of the model was tested through leave-one-out cross-validation, giving a Q(2)(CV) of 0.605 (Q(2)(CV) improves to 0.721 after removing two outliers). An external validation set was also used to give an account for the modelling capability, and the results agreed with the ones obtained from a 3D methodology previously applied to this series of compounds. The method showed to be a potential tool for predicting new drug-like compounds, as exemplified by calculating the activities of two new proposed congeners derived from the training set.