Although self-assessing one's knowledge is an integral part of a teacher's professional development, little is known about the accuracy of teacher candidates' self-assessments. In particular, not much attention has been paid to their self-assessment accuracy assessed on an individual level. Using measures from research on metacognition, we investigated the accuracy and bias of 513 mathematics teacher candidates who were tested in and self-assessed their professional knowledge in the three core domains of mathematical content knowledge, pedagogical content knowledge and pedagogical knowledge. In addition, we examined the consistency of self-assessment accuracy across the three knowledge domains and its relation to personal characteristics such as cognitive ability, personality traits and occupational self-efficacy. Results showed that in all three domains of professional knowledge, most teacher candidates were either over- or underconfident in their knowledge and unaware of their strengths and weaknesses. Cognitive ability reduced the risk of being overconfident, whereas self-efficacy was predictive of overconfidence. The Big Five personality traits were not systematically connected to self-assessment accuracy, and variance in self-assessment accuracy is still to be explained by further research. We address shortcomings of measures that are commonly used in self-assessment research and advocate for the use of metacognitive methods in future studies that focus on teacher self-assessments.