The relative hydration free energies of a series of organic molecules were calculated from molecular dynamics (MD) simulations using an extrapolation method in combination with soft-core potential scaling. This technique consists in first generating one single long trajectory using an unphysical reference state and in using afterward this trajectory for the estimation of the free energy difference between several molecules, First, we investigated the accuracy of the method for the deletion of small functional groups. A trajectory from an MD simulation of pyrogallol (1,2,3-trihydroxybenzene) was used to calculate by extrapolation the free energy changes for the mutations of pyrogallol, catechol, and phenol to benzene in water and in vacuo. The results were compared to those obtained by thermodynamic integration, to experimental data, and to values calculated using a semiempirical method. In a second step, increasingly larger mutations were studied in order to investigate the limitations of the method. The influence of various simulation parameters (choice of the unphysical reference state, simulation length. soft-core scaling parameter cr) on the final free energy values was examined. Benzene derivatives with hydroxyalkyl and/or bulky functional groups of increasing sizes were mutated into benzene. The results for simulations both in water and in vacuo were compared to the free energy results obtained by thermodynamic integration and to the experimental values of similar molecules, The results showed that for small mutations (deletion of functional groups with up to three atoms) the extrapolation method is reliable. However. the free energies calculated for the deletion of larger functional groups showed different accuracy levels depending on the chosen simulation parameters. For the largest mutations the thc thermodynamic integration method also showed convergence problems. This study therefore demonstrated the usefulness of the extrapolation method for molecules of similar size, but showed the difficulties of obtaining reliable results for molecules that substantially differ from each other.