Based on experimental research on helping and survey findings on drunk driving intervention, we selected 28 variables for study that might relate to informal intervention in a drunk driving situation. We tested six hypotheses by categorizing the 28 variables into personal (12), relationship (4), situational (5), and individual response (7) aspects. Bivariate correlations and multivariate logistic regression analyses were performed between these 28 predictor variables and intervention for 303 college students (192 females and 111 males). Three personal, one relationship, three situational, and four individual responses in situation variables were correlated significantly with intervention. Four significant predictors emerged in the stepwise logistic regression equation: knowing the driver well; having a conversation that encouraged the respondent to intervene; perceiving the driver as needing help; and the respondent's feeling able to intervene. All hypotheses received at least some confirmation in these results, but noteworthy discrepancies from expectations emphasize the need to appreciate differences between studies of helping behavior in the laboratory and this natural form of helping behavior.