Dual-process models propose that addictive behaviors are determined by an implicit, impulsive system and an explicit, reflective system. Consistent with these models, research has demonstrated implicit affective associations with alcohol, using the Implicit Association Test (IAT), that predict unique variance in drinking behavior above explicit cognitions. However, different IAT versions have been used to measure implicit affective associations with alcohol, and the present study sought to determine which of these IAT variants showed the highest validity and internal consistencies. in total, 4800 participants completed one of six IAT versions via the Internet: a bipolar IAT (i.e., positive vs. negative), a unipolar positive IAT (i.e., positive vs. neutral), or a unipolar negative IAT (i.e., negative vs. neutral) with general positive and negative stimuli or with positive and negative alcohol-related affective states. While the alcohol-related affective bipolar and unipolar positive IAT versions and the general affective bipolar and unipolar positive IAT versions showed comparable internal consistencies, somewhat lower internal consistencies were found for the unipolar negative IAT versions. Further, alcohol-related affective IAT variants were more strongly related to explicit measures than general affective IAT versions. Also, alcohol-related and general affective bipolar and unipolar positive IAT variants were related to drinking behavior, but not unipolar negative IAT variants. Finally, the bipolar alcohol-related affective IAT, the unipolar alcohol-related positive IAT and the unipolar general positive IAT predicted drinking behavior above explicit measures. Overall, the bipolar alcohol-related affective IAT outperformed all other IAT variants with respect to its relationship with explicit measures and drinking behavior.