An evaluation of different bottom friction parameterizations has been performed with a nonlinear barotropic shallow water model. The model was set up for the Baltic Sea and forced with meteorological data during 1 year. The simulations were validated using hourly observations from 38 tide gauge stations around the Baltic. A new parameterization with friction proportional to the scalar product between the flow velocity and the bottom slope was tested and compared with parameterizations using Rayleigh friction, Ekman friction, Laplacian viscosity, or quadratic bottom friction. The new parameterization consistently gave both a smaller error and a better correlation with the observations. The average error with slope friction was 5.0 cm, as compared with 5.2 cm using quadratic bottom friction, the best of the other parameterizations. The correlation using slope friction was 0.92, as compared with 0.91 using quadratic bottom friction. The results were slightly improved by combining slope friction and Rayleigh friction.