Employing data from a 1987 shelter survey of 1260 homeless adults in New York City, multivariate models of emergency room (ER) use are developed which include an array of risk factors for visiting a hospital ER including health and mental health problems, victimization and injuries. The study's primary goal is to identify factors that predict ER use in this population. Multivariate logistic and linear regression models were tested separately for men and women predicting three outcomes: any use of the ER during the past 6 months, use of the ER for injuries vs all other reasons (given any ER use), and the number of ER visits (given any ER use). Lower alcohol dependence, health symptoms and injuries were strong predictors for both men and women; other significant predictors differed markedly by gender. Both models were highly significant and produced strikingly high risk profiles. A high prevalence of victimization and injuries underlies ER use among the homeless. Based upon the findings, we recommend expanded health and victim services as well as preventive measures. Until primary care becomes available for this population, we advise against policies that discourage ER use by the homeless.