We examined whether students were accurate in predicting their test performance in both low-stakes and high-stakes testing contexts. The sample comprised U.S. high school students enrolled in an advanced placement (AP) statistics course during the 2017-2018 academic year (N = 209; M[subscript age] = 16.6 years). We found that even two months before taking the AP exam, a high stakes summative assessment, students were moderately accurate in predicting their actual scores (K[subscript weighted] = 0.62). When the same variables were entered into models predicting inaccuracy and overconfidence bias, results did not provide evidence that age, gender, parental education, number of mathematics classes previously taken, or course engagement accounted for variation in accuracy. Overconfidence bias differed between students enrolled at different schools. Results indicated that students' predictions of performance were positively associated with performance in both low- and high-stakes testing contexts. The findings shed light on ways to leverage students' self-assessment for learning.