Participants (n = 48) deprived of sleep for 29-50 hr, in comparison with controls (n = 45), underestimated their performance on logical reasoning and Raven's matrices. Such caution may ameliorate adverse practical consequences of sleep loss. In contrast, although sleep loss participants were more suggestible on the Gudjonsson Suggestibility Scale (G. H. Gudjonsson, 1984, 1987), they maintained confidence in their suggestible responses and were inaccurate when responding with the highest rating of confidence. This indicates that the increased suggestibility is internalized and is due to a cognitive deficit rather than to compliance. Eyewitness confidence-accuracy correlations were low but usually significant and were lowest after 47-50 hr of sleep loss. Repetition of leading questions led to increases in confidence for suggestible responses (with no interaction with sleep loss) but not for nonsuggestible responses, indicating a problem for jurors' evaluations of practiced testimony.