Williams's (2005) study on "learning without awareness" and three subsequent extensions (Faretta-Stutenberg & Morgan-Short, 2011; Hama & Leow, 2010; Rebuschat, Hamrick, Sachs, Riestenberg, & Ziegler, 2013) have reported conflicting results, perhaps in part due to differences in how awareness has been measured. The present extension of Williams (2005) addresses this possibility directly by triangulating data from three awareness measures: concurrent verbal reports (think-aloud protocols), retrospective verbal reports (postexposure interviews), and subjective measures (confidence ratings and source attributions). Participants were exposed to an artificial determiner system under incidental learning conditions. One experimental group thought aloud during training, another thought aloud during training and testing, and a third remained silent, as did a trained control group. All participants were then tested by means of a forced-choice task to establish whether learning took place. In addition, all participants provided confidence ratings and source attributions on test items and were interviewed following the test. Our results indicate that, although all experimental groups displayed learning effects, only the silent group was able to generalize the acquired knowledge to novel instances. Comparisons of concurrent and retrospective verbal report data shed light on the conflicting findings previously reported in the literature and highlight important methodological issues in implicit and explicit learning research.