Two studies examined the accuracy and differentiation of 4–5-yearolds‘, 8–9-year-olds’, and undergraduates' predictions of the preferences of peers and nonpeers. In Study 1 each subject was presented with separate arrays of snacks, meals, and activities depicted on cards and were asked to select their own preferences and the preferences of peers and nonpeers (“grown-ups” for the children, and “4- to 5-year-olds” for the undergraduates). In Study 2 each subject selected his or her own preference, the preference of peers, and the preferences of both older nonpeers (“grown-ups”) and younger nonpeers (“2-year-olds”). For all age groups, including 4–5-year-olds: (1) the preference predictions differentiated peers from nonpeers, as well as older nonpeers from younger nonpeers; (2) it was very rare for a subject to select his or her own preferences for the preference predictions of both peers and nonpeers. There were no consistent developmental differences either in the tendency to select one's own preferences when predicting the preferences of others or in the tendency to differentiate predictions for peers and nonpeers. In contrast, there was a clear developmental increase in predictive accuracy, with 4–5-year-olds being relatively inaccurate in predicting the preferences of nonpeers. The inadequacy of constructs such as “assumed similarity” and “egocentrism” as explanations for the general accuracy in predicting peers' preferences and the 4–5-year-olds' inaccuracy in predicting nonpeers' preferences is discussed. Possible alternative variables underlying developmental increases in judgmental accuracy, such as “social reference,”“self reference,” and “social category knowledge,” are then proposed.