Research indicates that learning from erroneous examples (EE) is superior to correct examples because errors may provide students with a stimulus to spontaneously produce more self-explanations, leading to better learning outcomes. However, because most studies were conducted in individual settings, it remains an open question whether the advantage of EE could be reproduced in collaborative settings. Therefore, this study used a 2 × 2 factorial design varying in example types (correct vs. erroneous examples) and learning settings (individuals vs. collaboration). An experimental study of a 2-week course on subtraction for third grade students was conducted (N = 109). Results indicated that collaborative learning may enhance the transfer knowledge of subtraction. The improvements were still maintained over 2 weeks. In addition, collaboration also facilitated students' confidence in handling errors. However, there was no significant difference in promoting isomorphic and transfer knowledge between learning from correct and erroneous examples. Additionally, this study also explored the absence of the intended benefit of erroneous examples from a cognitive load perspective.