Because animations can depict situational dynamics explicitly, they have the potential to help learners build coherent, high-quality mental models of complex change processes. Further, "interactive" animations provide opportunities for learners to deal with available information selectively and so avoid excessive processing demands. However, to be instructionally effective, the selected subsets of information must have high domain and task relevance. Approaches used by domain novices to interrogate an interactive animation of a complex dynamic system as they prepared for a subsequent prediction task were explored. Subjects searched the animation in order to learn generalizations upon which to base their predictions. Spatial and temporal strategies employed tended to be narrowly focused upon individual graphic features or localized groups while broader relational aspects required for coherence were neglected. The findings suggest that in order to build satisfactory mental representations from interactive animations, learners may require specific guidance regarding search strategies and targets.