Off-site surveillance involves using financial ratios to identify banks likely to develop safety-and-soundness problems. Bank supervisors use two tools to flag developing problems: supervisory screens and econometric models. Despite the statistical dominance of models, supervisors continue to rely heavily on screens. We use data from the 1980s and 1990s to compare, once again, the performance of the two approaches to off-site surveillance. Our study explicitly addresses supervisors' criticisms of econometric models. In particular, we offer a new econometric model - one designed to forecast downgrades in supervisory ratings - that is more forward-looking than existing models. As in earlier comparisons, econometric models consistently outperform supervisory screens for our sample. These results do not, however, suggest that screens should be dropped from the surveillance toolbox. When abrupt changes in the causes of bank failures and CAMEL downgrades occur, supervisors can modify their screens long before models can be revised to reflect the new conditions. We conclude that both screens and models add value in off-site surveillance, but that supervisors should rely more heavily on econometric models in the future than they have in the past.