Educational games and tutors provide conflicting approaches to the assistance dilemma, yet there is little work that directly compares them. This study tested the effects of game-based and tutor-based assistance on learning and interest. The laboratory experiment randomly assigned 105 university students to two versions of the educational game Policy World designed to teach the skills of policy argument. The game version provided minimal feedback and imposed penalties during training while the tutor version provided additional step-level, knowledge-based feedback and required immediate error correction. The study measured students' success during training, their interest in the game, and posttest performance. Tutor students were better able to analyze policy problems and reported higher level of competence which in turn affected interest. This suggests that we can improve the efficacy and interest in educational games by applying tutor-based approaches to assistance.