Objective: To identify predictors of error in estimated blood alcohol concentration (eBAC) in a sample of bar patrons. Method: Six hundred sixty-six patrons (43.6% female) were randomly sampled from 32 bars. Patrons were asked to provide a breath sample into a handheld breath alcohol concentration test unit upon entrance and exit from the bar. Patrons also completed a brief survey at entrance and exit. For analyses, this sample was stratified by whether patrons consumed alcohol before attending the bar. Estimates of BAC were calculated using Matthews and Miller's formula (1979). A three-category dependent variable was created based on the estimation accuracy of eBAC relative to breath alcohol concentration: accurate (within .02), underestimate of BAC, and overestimate of BAC. Results: Of those that drank before arriving at the bar, 29% of eBACs were accurate, 32.3% were underestimates, and 38.8% were overestimates. For those who drank only at the bar, 42.0% were accurate, 20.8% were underestimates, and 37.7% were overestimates. Among those who drank before attending the bar, the number of drinks consumed before attending the bar was significantly related to eBAC underestimate. Among those who drank only at the bar, predictors of overestimate included being female, drinking more, and drinking longer. Conclusions: The accuracy of eBAC is poor at best. In an earlier study of parties, eBACs were often underestimated; in the bar setting, eBACs were often overestimated. More research is needed to understand the role of setting on eBAC calculations.