The usefulness of financial accounting ratios has long been documented in statistical models of business failure prediction. However, behavioral studies to date have reported inconsistent results, with prediction accuracy ranging from 41 percent to 93 percent. These studies have also presented varying explanations for the inconsistencies. Data from 33 previous experimental results were analyzed in this meta-analysis to determine the moderating effects of differences in task properties on failure prediction accuracy. As expected, all task properties were found to have a significant moderating effect on prediction accuracy. These findings have important implications for future research and practice in the prediction of business failure and similar tasks.