Adopting a pretest-posttest experimental design with repeated measures, this study examined the effects of three types of game-based learning supports in the form of modeling on knowledge development that contributed to successful math problem solving and students' perceived game flow. Forty-one sixth-grade students participated in the study and played a 3D architecture game that aims to promote mathematical conceptual understanding and problem solving skills, and presented with different types of learning supports (i.e., conceptual knowledge only, procedural knowledge only, and the combination of the two). No significant effect of the types of supports was found on participants' perceived game flow. However, there was a significant impact of support type on participants' performance on two posttests. The results indicated that the learning support with procedural knowledge only, compared with the learning support with both conceptual and procedural knowledge as well as that with conceptual knowledge only, was significantly more effective in promoting students' knowledge acquisition related to mathematical problem solving.