In the current situation where no consensus has been reached regarding the output variables and input factor variable sets used in evaluating bank productivity, this study synthesized the various variable sets used in past bank productivity-related research, classified them into eight variable sets, evaluated productivity, and analyzed the differences in productivity evaluation values among variable sets, i.e., sensitivity. Forty-nine branch offices of a commercial bank were selected as evaluation targets, and DEA (Data Envelopment Analysis) was used as the productivity evaluation method. The absolute efficiency value results showed statistically significant differences among variable sets in most cases, while the relative ranking results did not show significant differences among variable sets. In the analysis of banks' average behavioral patterns, generally consistent results were obtained regardless of the variable set selected for analysis. No meaningful correlations were found with other comprehensive performance measures such as return on total assets or profit per employee, which implies that evaluation based on a single performance measure entails considerable risk. It was found that efficiency values using only financial variables presented in financial statements tended to be more stable compared to results from other variable sets that included physical quantity variables, suggesting that information provided by financial statements can be usefully employed in evaluating bank productivity.