Abstract Sex and ability differences in math self-efficacy and prediction accuracy were examined in 38 gifted and 38 average ability fifth graders. Utilizing a 5-point certainty scale, subjects made predictions about their ability to solve each problem on a 20-item math task. After predictions were made, participants completed the math task in full. As expected, a significant ability effect in favor of gifted subjects was obtained for self-efficacy level (number of items out of 20 with certainty ratings above 3), self-efficacy strength (average of certainty ratings), and prediction accuracy (number of hits for items with certainty ratings above and below 3). Significant sex differences were found on the self-efficacy strength index; regardless of ability level, males showed more self-efficacy strength for the math task than was demonstrated by their female counterparts. Prediction accuracy analyses of success expectations revealed that gifted subjects made significantly fewer over-estimations and that males tended to overestimate more than females. Average ability subjects' low self-efficacy ratings were found to be significantly more accurate than those made by gifted subjects. Findings were discussed in relation to educational implications and directions for continued investigation.