Despite the potential benefits of implicit measures over self-report measures, they are rarely used in real-world contexts to predict behavior. Two potential reasons are that (a) traditional implicit measures typically show low predictive validity and (b) the practical utility of implicit measures has hardly been investigated. The current studies test the practical utility of a new generation of implicit measures for predicting drunk driving. Study 1 (N = 290) examined whether an implicit measure of beliefs about past drunk driving (i.e., the Past Driving Under the Influence Implicit Association Test; P-DUI-IAT) retrospectively predicts drunk driving in driving school students, a population for which this measure could have applied value. Study 1 also explored whether P-DUI-IAT scores prospectively predicted drunk driving over six months. Due to the low number of offenders, however, Study 1 had low statistical power to test this latter question. In Study 2 (N = 228), we therefore examined the utility of the P-DUI-IAT and a new variant of this test (i.e., the Acceptability of Driving Under the Influence Implicit Association Test; A-DUI-IAT) to prospectively predict drunk driving in an online sample with a high number of offenders. Results from Study 1 show that the P-DUI-IAT predicts self-rated past drunk driving behavior in driving school students (ORs = 3.11–6.12, ps < .043, 95% CIs = [1.11, 37.69]). Results from Study 1 do not show evidence for utility of the P-DUI-IAT to prospectively predict self-rated drunk driving. Results from Study 2, on the other hand, show strong evidence for the utility of both implicit measures to prospectively predict self-rated drunk driving (ORs = 3.80–5.82, ps < .002, 95% CIs = [1.72, 14.47]). Although further applied research is necessary, the current results could provide a first step towards the application of implicit measures in real-world contexts.