Abstract Prediction of the retention times of steroids in two liquid chromatographic systems, i.e., straight- and reversed-phase, has been modeled by means of different quantitative structure-retention relationship (QSRR) methods. The 3D-QSRR descriptors were generated by a process including semiempirical structure optimization, structure alignments, and electronic 3D field descriptions by the GRID method or by a novel method based on summation of the electronic charges at grid points ( Q-field). The compound data set was split into a calibration set and a test set using the self-organizing feature map (SOFM) technique. The correlation between the molecular descriptors and the retention time of the compounds in the calibration set was established using the partial least squares (PLS) method. Optimal predictive models were generated by a cross-validation procedure. These models were then used to predict the retention times of the compounds in the independent test set. A further check on the validity of the models was obtained by back-projection of the most important variables of the models onto the molecular structure. In addition, the predictive capability of variable-reduced models was examined. The various 3D-QSRR models generated gave for the reversed-phase system Q prediction 2 values in the range 0.60–0.79 for the test set, the corresponding Q prediction 2 values for the straight-phase system being in the range 0.42–0.75. In the former case, variable-reduced models resulted in considerably better predictions, although these were not as good as for those models obtained by means of classical physical–chemical descriptors.