In a follow-up study to Dunning, Griffin, Milojkovic, and L. Ross (1990), which had investigated the phenomenon of overconfidence in social prediction, two samples of first-year undergraduates were invited to make predictions about their own future responses (and, in the case of Sample 2, also those of their roommates) over the months ahead. These predictions were accompanied by confidence estimates and were evaluated in the light of actual responses reported later by the subjects in question. The primary finding was that self-predictions, like social predictions, proved to be consistently overconfident. As in Dunning et al., moreover, overconfidence could be traced to two sources. First, expressions of particularly high confidence rarely proved to be warranted; as confidence increased, the gap between accuracy and confidence widened. Second, predictions that went against relevant base rates yielded very low accuracy in the face of relatively unattenuated confidence levels. The implications of these results are discussed, and one potentially important underlying mechanism--the failure to make adequate inferential allowance for the uncertainties of situational construal--is proposed for further research.