To evaluate the level of agreement between two sets of criteria, the Thrombolysis in Myocardial Infarction (TIMI) criteria and investigator-developed criteria, for identifying bleeding events in patients who had undergone a percutaneous coronary intervention (PCI) and to measure length of hospital stay (LOS) as a surrogate marker of bleeding severity. Retrospective chart review. Two university-affiliated medical centers. Four hundred twenty-two consecutive patients who had undergone PCI from December 1, 2001-June 30, 2002. Data were collected on the number of bleeding events that occurred within 1 week after PCI (limited to one event/patient) and on LOS. Bleeding was assessed by TIMI criteria and by investigator-developed criteria. Bleeding according to TIMI criteria included intracranial hemorrhage, spontaneous hematuria or hematemesis, or decreases in hemoglobin level. Bleeding according to investigator-developed criteria included intracranial, retroperitoneal, intraocular, or clinically overt bleeding without a specified decrease in hemoglobin level. Agreement between criteria was assessed by means of the kappa statistic. Of the 422 patients, 23 (5%) experienced TIMI-defined bleeding events and 229 (54%) investigator-defined bleeding events. A kappa value of 0.09 (95% confidence interval 0.06-0.13) indicated a poor level of agreement between the two sets of bleeding criteria. The effect of this discrepancy on LOS was 66 fewer days of care when TIMI criteria were applied versus investigator-developed criteria in the 206 patients who experienced bleeding events that met investigator criteria (total LOS 645 days) but not TIMI criteria (total LOS 579 days). Bleeding assessment with use of TIMI criteria versus investigator-developed criteria yielded discrepant bleeding event rates and LOS, making it difficult to accurately compare bleeding rates and consequences across clinical trials and in practice. Consensus bleeding criteria are needed for applications in clinical evaluations of antithrombotic agents.