The present study examined the cognitive factors that influence children's physical science learning from a multimedia instruction. Using a causally coherent text and visual models, we taught 4th- and 7th-grade children about the observable and molecular properties of the three states of water. We manipulated whether the text was read by a tutor (which supports simultaneous encoding of the verbal and visual information, i.e., "temporal contiguity") or whether children read the text on their own (which supports self pacing and interpretation of the information). Children in each condition received either static or dynamic graphics. Results showed that, regardless of the type of graphics, children demonstrated the greatest learning gains when the text was read to them by a tutor. This effect was more pronounced for the younger children. Thus, conditions that promote integration of verbal and visual information may provide the greatest support to children's learning from a causally coherent multimedia science lesson.