Logistic regression was used to compare the predictive value of alcohol consumption variables on casualty occurrence in probability samples of two diverse emergency room (ER) populations who were breathalyzed and interviewed at the time of ER admission: San Francisco General Hospital (N = 1459) and four hospitals representative of a nearby California county (N = 1756). Previous analyses suggested differences in the association of alcohol consumption and injury in these two samples. Age, breathalyzer reading and feeling drunk at the time of the event were consistent predictors of injury status in both samples while breathalyzer reading, quantity and frequency of usual drinking and place of injury occurrence (workplace vs. home) were consistent predictors of alcohol-related injuries (self-reported drinking 6 h prior to event).