Many real-world eyewitnesses are under the influence of alcohol either at the time of the crime, the interview, or both. Only recently has empirical research begun to examine the effects of alcohol on witness memory, yielding mixed results. The present study tested the importance of state-dependent memory in the context of alcohol’s effects on encoding versus retrieval of a witnessed event, while simultaneously informing real-world investigative practices: Should witnesses sober up before an interview? Participants (N = 249) were randomized to a control, placebo, or alcohol condition at encoding and to either an immediate retrieval condition (in the same state) or a 1-week delay control, placebo, or alcohol retrieval condition. They recalled a witnessed mock crime using open ended and cued recall formats. After a delay, witnesses intoxicated at both encoding and retrieval provided less accurate information than witnesses in sober or placebo groups at both times. There was no advantage of state-dependent memory but intoxicated witnesses were best when recalling immediately compared to 1 week later (sober, placebo, or reintoxicated). Findings have direct implications for the timing of intoxicated witnesses’ interviews such that moderately intoxicated witnesses may not benefit from a sobering delay but rather, should be interviewed immediately.