Besides learners' awareness of their knowledge, a growing number of studies also emphasise the importance of teachers' awareness of how well their students perform to adjust their teaching strategies accordingly. Therefore, proposing a multi-layered metacognitive regulatory model in teaching first, we investigated whether estimation type, item difficulty, and class performance affect teachers' judgment accuracies ([JAs], i.e., score estimations). Teachers (N=38) of 86 classes made item-by-item and overall estimations of their classes' test scores (N=2608 sixth-graders native in Turkish) at a PISA-equivalent mathematics test that was developed in the earliest phase of the current long-term research project. The results showed that teachers' item-by-item estimations were below their classes' actual performance, unlike their overall estimations. Teachers of low-performance classes were less accurate than those of high-performance classes. These teachers also showed the clearest underestimation for the easy questions, whereas teachers of high-performance classes overestimated their classes' scores for the difficult questions. This dissociation implied that the teachers 'must have' primarily used their perceptions about their classes (e.g., classes' existing performance) as a mnemonic judgment cue rather than item difficulty as an external cue when making their score estimations. The implications of the results were discussed in the light of existing literature and suggestions for prospective research were given.