The article reviews the MM/PBSA and MM/GBSA methods, which combine molecular mechanics with Poisson-Boltzmann or generalized Born and surface area continuum solvation to estimate ligand-binding affinities. These methods are widely used due to their modular nature and the absence of the need for a training set, but they suffer from several crude approximations, such as the lack of conformational entropy and information about water molecules in the binding site. The authors discuss the precision and accuracy of these methods, their variants, and attempts to improve them using more accurate approaches like quantum mechanics, polarizable force fields, or improved solvation models. They find that while MM/PBSA can reproduce and rationalize experimental findings, it is not accurate enough for predictive drug design. The methods' performance varies significantly depending on the tested system, and attempts to improve them often lead to inconsistent results. The article concludes that MM/PBSA may be useful for post-processing docked structures or rationalizing observed differences but is not suitable for predictive drug design.The article reviews the MM/PBSA and MM/GBSA methods, which combine molecular mechanics with Poisson-Boltzmann or generalized Born and surface area continuum solvation to estimate ligand-binding affinities. These methods are widely used due to their modular nature and the absence of the need for a training set, but they suffer from several crude approximations, such as the lack of conformational entropy and information about water molecules in the binding site. The authors discuss the precision and accuracy of these methods, their variants, and attempts to improve them using more accurate approaches like quantum mechanics, polarizable force fields, or improved solvation models. They find that while MM/PBSA can reproduce and rationalize experimental findings, it is not accurate enough for predictive drug design. The methods' performance varies significantly depending on the tested system, and attempts to improve them often lead to inconsistent results. The article concludes that MM/PBSA may be useful for post-processing docked structures or rationalizing observed differences but is not suitable for predictive drug design.