Students’ Judgments of Learning (JOLs) are often inaccurate: students often overestimate their future test performance. Because of the consequences that JOL inaccuracy can have for regulating study activities, an important question is how JOL accuracy can be improved. When learning texts, JOL accuracy has been shown to improve through ‘generation strategies’, such as generating keywords, summaries, or concept maps. This study investigated whether JOL accuracy can also be improved by means of a generation strategy (i.e., completing blank steps in the examples) when learning to solve problems through worked example study. Secondary education students of 14–15 years old (cf. USA 9th grade) either studied worked examples or completed partially worked examples and gave JOLs. It was found that completion of worked examples resulted in underestimation of future test performance. It seems that completing partially worked-out examples made students less confident about future performance than studying fully worked examples. However, this did not lead to better regulation of study.