Although educational computer games have been recognized as being a promising approach, previous studies have indicated that, without supportive models, students might only show temporary interest during the game-based learning process, and their learning performance is often not as good as expected. Therefore, in this paper, a two-tier test approach is proposed for developing educational computer games. To evaluate the performance of the proposed approach, a role-playing game was developed and an experiment was conducted on the "migratory bird identification" unit of an elementary school natural science course. Two classes of fifth graders participated in the experiment. One class was the experimental group who learned with the proposed game-based learning approach. The other class was the control group who learned with the conventional e-learning approach. The experimental results showed that the proposed approach not only significantly promoted the students' learning motivation, but also improved their learning achievements. Moreover, it was found that the students who learned with the educational computer game showed significantly higher technology acceptance degree than those who learned with the conventional e-learning approach. Accordingly, it is concluded that the proposed digital game-based learning approach is effective.