The number of stream restoration projects has increased dramatically during the last decade, especially in the New York City watershed region, where stream management to improve water quality is a high priority, and where the NYC Department of Environmental Protection and the Greene County Soil and Water Conservation District have partnered to develop a set of restoration demonstration projects. In this paper, the effectiveness of stream restoration projects in reducing stream bank erosion in the Batavia Kill watershed (Greene County, New York) is evaluated. This evaluation is based on a multivariate regression model to relate stream bank erosion rates to various explanatory variables including instruments representing geomorphological characteristics, flow conditions, rainfall conditions, temperature, the vegetation index, soil erodibility, and sediment characteristics. The general to specific approach is used to specify the regression model. A range of statistical tests is applied to check the model accuracy and the validity of the regression model. The results of these tests show that the stepwise regression model accurately predicts stream bank erosion rates on the Batavia Kill stream. The regression model is then applied on the project reaches, assuming there was no stream restoration to predict the stream bank erosion. It is found that the measured erosion on the restored reaches is much smaller than predicted erosion in the "without restoration" case, which means that the effectiveness of stream restoration in reducing bank erosion in the Batavia Kill watershed is significant.