The accurate prediction of ligand–receptor interaction energies by molecular modelling methods remains challenging. Predicting and understanding the sequence-selectivity of DNA minor groove-binding ligands constitutes a particularly interesting and potentially valuable aspect of this. Here, we use experimental data on the binding of Hoechst 33258 to the minor groove of various A/T-rich DNA duplexes to evaluate the reliability of a popular class of molecular modelling methods based on the energetic analysis of molecular dynamics (MD) simulations. We examine how performance depends on the use of explicit versus implicit solvent models, on the use of generalised Born versus Poisson–Boltzmann models to evaluate solvent-associated energetic terms, and the use of normal mode analysis to evaluate entropic factors. Quantitatively evaluating many different combinations of methodologies, we find that the most reliable results are obtained when the MD simulations are performed in explicit solvent, when the data is processed using the MM-PB/SA approach, and when normal mode analysis is used to estimate configurational entropy changes.