The current meta-analysis quantifies the average effect of worked examples on mathematics performance from elementary grades to postsecondary settings and to assess what moderates this effect. Though thousands of worked examples studies have been conducted to date, a corresponding meta-analysis has yet to be published. Exclusionary coding was conducted on 8033 abstracts from published and grey literature to yield a sample of high quality experimental and quasi-experimental work. This search yielded 43 articles reporting on 55 studies and 181 effect sizes. Using robust variance estimation (RVE) to account for clustered effect sizes, the average effect size of worked examples on mathematics performance outcomes was medium with g = 0.48 and p = 0.01. Moderators assessed included example type (correct vs. incorrect examples alone or in combination with correct examples), pairing with self-explanation prompts, and timing of administration (i.e., practice vs. skill acquisition). The inclusion of self-explanation prompts significantly moderated the effect of examples yielding a negative effect in comparison to worked examples conditions that did not include self-explanation prompts. Worked examples studies that used correct examples alone yielded larger effect sizes than those that used incorrect examples alone or correct examples in combination with incorrect examples. The worked examples effect yields a medium effect on mathematics outcomes whether used for practice or initial skill acquisition. Correct examples are particularly beneficial for learning overall, and pairing examples with self-explanation prompts may not be a fruitful design modification. Theoretical and practical implications are discussed.