Two experiments examined visual attention distribution in learning from text and pictures. Participants watched a 16-step multimedia instruction on the formation of lightning. In Experiment 1 (N=90) the instruction was system-paced (fast, medium, slow pace), while it was self-paced in Experiment 2 (N=31). In both experiments the text modality was varied (written, spoken). During learning, the participants' eye movements were recorded. Results from both experiments revealed that learners spent more time studying the visualizations with spoken text than those with written text. In written text conditions learners consistently started reading before alternating between text and visualization; moreover, they spent more time reading the text than inspecting the visualizations. While in Experiment 1 additional time that was made available in conditions with a slow or medium instruction pace was spent inspecting visualizations, in Experiment 2 longer learning times resulted from reading the text more intensively. With respect to learning outcomes (retention, transfer, and visual memory) Experiment 1 revealed an effect of text modality for visual memory only. In Experiment 2 no modality effects were found. Instruction pace was hardly related to learning outcomes. Overall, the results confirm prior findings suggesting that the distribution of visual attention in multimedia learning is largely guided by the text. (Contains 4 tables and 4 figures.)