We describe an experiment to assess the effectiveness of a fish diversion system that utilizes high-frequency sound at the Annapolis Tidal Generating Station, No a Scotia. Canada. during the fall of 1999. A band-limited, random-noise signal, with most of the energy focused between 122 and 128 kHz, was projected into the turbine forebay during randomly selected generating cycles. The effectiveness of the diversion system was assessed by monitoring fish passage through the turbine and two adjacent fishways. During, the study, fish representing 27 taxa were captured. For the 11 species with sufficient data, we modeled the rate of passage as a function of the sampling site and the on/off status of the diversion system and compared models with and without a set of environmental variables. The environmental component of the model was highly significant for all 11 species. When the environmental variables were removed from the models, the standard errors of the diversion coefficients increased, and between-site comparisons showed that factors other than the on/off status of the diversion system were affecting the effectiveness estimates. Model coefficients were estimated using maximum likelihood, assuming Poisson or extra-Poisson error distributions. The catches of all 11 species were overdispersed, and the statistical significance of the effectiveness estimates was overestimated when a Poisson error distribution was assumed. We conclude that the diversion system reduced passage through the turbines for members of the genus Alosa but that it was not effective for the other eight specics. With the diversion system activated, the rates of passage of American had A. sapidissima and alewife A. pseudoharengus through the turbine tube decreased by 42% and 48%, respectively. For blueback herring A. aestivalis, no diversion effect was detected when all catches were included in the analysis However, a 49% decrease in the rate of passage was detected alter the tidal cycles when the three largest catches were trimmed from the data