Erroneous examples--step-by-step problem solutions with one or more errors for students to find and fix--hold great potential to help students learn. In this study, which is a replication of a prior study (Adams et al. 2014), but with a much larger population (390 vs. 208), middle school students learned about decimals either by working with interactive, web-based erroneous examples or with more traditional supported problems to solve. The erroneous examples group was interactively prompted to find, explain, and fix errors in decimal problems, while the problem-solving group was prompted to solve the same decimal problems and explain their solutions. Both groups were given correctness feedback on their work by the web-based program. Although the two groups did not differ on an immediate post-test, the erroneous examples group performed significantly better on a delayed test, given a week after the initial post-test (d?=?0.33, for gain scores), replicating the pattern of the prior study. Interestingly, the problem solving group reported liking the intervention more than the erroneous examples group (d?=?0.21 for liking rating in a questionnaire) and found the user interface easier to interact with (d?=?0.37), suggesting that what students like does not always lead to the best learning outcomes. This result is consistent with that of desirable "difficulty" studies, in which a more cognitively challenging learning task results in deeper and longer-lasting learning.