Rescoring of Center for Occupational and Professional Assessment objective-referenced tests is decided largely by content experts selected by client organizations. A few of the test items, statistically flagged for review, are not rescored. Some of this incongruence could be due to the use of the biserial correlation (r-biserial) as an item-discrimination index. Item-flagging outcomes generated using the present item-discrimination index, r-biserial, were compared with those generated using phi/phi max and phi. The analysis sample consisted of responses of 10,702 candidates to 1,794 items taken from 23 different test forms. Major findings suggest that the incongruent items, initially flagged using r-biserial, tended also to be flagged when either phi/phi max or phi were employed. The item-rescore decisions made by the content experts were unsupported by the results, suggesting that their actions should be more closely monitored. Some implications for use of the three item-discrimination indices are also discussed. (Author)