Previous studies suggest that cues predicting the outcome of attentional shifts provide a measure of anticipatory alcohol-related attentional bias that is correlated with risky drinking and has high reliability. However, this is complicated by potential contributions of visual features of cues to reliability, unrelated to their predictive value. Further, little is known of the sensitivity of the bias to variations in cue-outcome mapping manipulations, limiting our theoretical and methodological knowledge: Does the bias robustly follow varying cue-outcome mappings, or are there automatic cue-related associative processes involved? The current studies aimed to address these issues. Participants performed variations of the cued Visual Probe Task (cVPT) in which cues were non-predictive; in which there were multiple cue pairs, used simultaneously and serially; and in which the cue-outcome mapping was reversed. The major findings were, first, that previously found reliability cannot be attributed to aspects of the cues not related to outcome-prediction; second, that reliability of the bias does not survive deviations from a simple, consistent cue-outcome mapping; third, that all predictive versions of the task showed a bias towards alcohol; fourth, that the bias did not simply follow awareness of the cue-outcome mapping; and finally, that only in the case of simultaneous multiple cue pairs, an association with risky drinking was replicated. The results provide support for the reliability of the anticipatory attentional bias for alcohol, suggest that relatively persistent associative processes underlie the bias in the alcohol context, and provide a foundation for future work using the cVPT.