Generally, the conformational equilibrium of flexible molecules changes upon solvation. The equilibrium distribution of conformers is determined by the potential of mean force for the relevant degrees of freedom. However, the determination of the potential of mean force by straightforward computer simulation is often impracticable owing to the presence of large energy barriers which prevent efficient sampling. A frequently chosen technique to overcome the problems associated with such barriers is the umbrella sampling technique. Using molecular dynamics simulation of 1,2-dichloroethane solvated in water as a model case, we have investigated the influence of the choice of the biasing potential in the umbrella sampling technique on the computation of the dihedral angle potential of mean force. The efficiencies of one- and multiple-window techniques are compared. For short simulation periods (25 ps) the latter technique performs better than the former. For longer simulation periods the performance difference is obscured by fluctuations caused by insufficient sampling of other slow processes occurring in the system, induced by the anisotropy of the periodic boundary conditions.