Numerous studies have found that banks exhaust scale economies at low levels of output, but most are based on the estimation of parametric cost functions which misrepresent bank cost. Here we avoid specification error by using nonparametric kernel regression techniques. We modify measures of scale and product mix economies introduced by Berger et al. (1987) to accommodate thenonparametric estimation approach, and estimate robust confidence intervals to assess the statistical significance of returns to scale. We find that banks experience increasing returns to scale up to approximately 0 million ofassets, and essentially constant returns thereafter. We also find that minimum efficient scale has increased since 1985.