Making errors is part of human nature, and it is thus important to know how to get the best out of them. Experimental evidence has shown that generating errors can enhance learning when these are followed by corrective feedback. However, little is known about the specific conditions and mechanisms that underlie this benefit of experiencing errors. This review aimed to shed some light on this type of learning. First, we highlight certain conditions that may influence errorful learning. These include the timing of corrective feedback, error types, learner awareness about errorful learning, motivation to learn the study material, differences in special populations (e.g., amnesia), incidental versus intentional encoding, the importance of selecting an appropriate final test procedure, whether the study material needs to be semantically related, and if it is necessary to recover the previous errors at the time of retrieval. We then consider four explanatory theories of errorful learning: (1) The Mediator Effectiveness hypothesis, (2) the Search Set theory, (3) the Recursive Reminding theory, and (4) the Error Prediction theory. According to these theories, two factors are decisive for observing the benefits of errorful learning: the level of a pre-existing semantic relationship between the study materials, and whether the error must be explicitly recovered on the final test. To conclude, we discuss some limitations of using a pretesting procedure to study errorful learning and we reflect on further research. This review brings us closer to understanding why experiencing errors confers a memory advantage.