Inspired by a paper at last year's PERC conference, in which Rebello [1] compared students' individual and cohort mean score estimations with their actual assessment scores, we present results of a study in which students in an introductory physics class were asked to predict their scores on two assessments, one delivered at the start of the course (pre-instruction) and one at the end of the course (post-instruction). Our results show that, pre-instruction, the academically strongest students tend to underestimate their score slightly, whereas the weakest overestimate their performance significantly, consistent with the findings of Rebello and demonstrating a well-known cognitive bias (the Dunning-Kruger effect). Post-instruction, we find that the ability of the original weakest quartile cohort to accurately predict their own assessment score has improved significantly, but a flux of students between quartiles from one assessment to the other reveals that the least able students continue to over-estimate their performance, but with a reduced mean discrepancy. We discuss the implications these results have for instruction and for development of enhanced metacognition amongst physics students.