The objective of this study is to evaluate the performance of several open boundary conditions applied to the Princeton ocean model. The focus is on passive open boundary conditions applied to the external mode, i.e., conditions that are applied when the mean flow at the open boundary is unknown and the values of the variables must be assumed or extrapolated from the interior solution. Three types of open boundary conditions (OBCs) are tested: (1) radiation conditions, (2) characteristic methods, (3) relaxation schemes. Numerical experiments are conducted in a zonal channel that include different forms of bottom topography. The experiments are designed to emphasize flow conditions dominated by direct wind forcing or wave radiation. Three sets of experiments are discussed: (1) forcing by a uniform, alongshelf wind stress; (2) the barotropic adjustment of an initial perturbation in the sea surface elevation; and (3) forcing by a traveling storm. The results of these experiments are compared with analytical solutions and the results of experiments using cyclic boundary conditions, or expanded domains. According to our results all radiation schemes of Orlanski's type perform poorly in either flows with strong nonlinear components or when the model response is dominated by the propagation of dispersive wave packets. The characteristic method recommended by Roed and Cooper <1987> provided reasonable responses in most cases, but we found problems in handling dispersive waves or variable wind forcing at the open boundaries. Although there were limitations in all the OBCs that we tested the best overall performances were for the conditions proposed by Flather <1976>, combined with a local solution approach proposed by Roed and Smesdtad <1984>, and the flow relaxation scheme developed by Martinsen and Engedahl <1987>. ¿ 1998 American Geophysical Union