Learner-generated drawing is a strategy that can improve learning from expository text. In this paper, a model of drawing construction is proposed and the experimental design tests hypotheses derived from this model. Fourth and sixth grade participants used drawing under three experimental conditions with two conditions including varying degrees of support. On a problem solving posttest, both supported drawing groups scored higher than the non-drawing Control group. Although the grade by condition interaction was not significant, there was a strong trend in this direction. When sixth grade participants were considered independently, participants in the most supported drawing condition also obtained higher problem solving scores than those who drew without support. There were no significant condition effects for fourth grade nor were there any significant effects on a multiple-choice recognition posttest. Results were consistent with hypotheses and are discussed in light of the proposed theoretical framework.