The authors examined effects of 3 metacognitive approaches and 1 control group on mathematical reasoning, conceptual errors, and metacognitive knowledge. The metacognitive approaches were (a) diagnosing errors (DIA), (b) improvement via self-questioning (IMP), and (c) a combined approach (DIA+IMP). Controls (CONT) received no metacognitive instruction. Participants were 115 9th-grade students in an Israeli junior high school studying linear functions and graph representations. Results indicated that students using the combined approach outperformed their peers in all measures of mathematical reasoning and metacognitive knowledge. The IMP students outperformed the DIA and CONT students in problem-solving skills and metacognitive strategy use, and the DIA students outperformed their peers in reducing conceptual errors. The authors discuss the practical and theoretical implications of these metacognitive approaches. (Contains 5 tables and 3 figures.)