The purpose of the experiment was to examine whether students better understand a science text when they are asked to self-generate summaries or to study predefined summaries. Furthermore, we tested the effects of verbal and pictorial summaries. The experiment followed a 2 × 2 design with representation mode (verbal vs. pictorial) and learning activity (self-generating vs. studying) as experimental factors. The main dependent variables were learning performance, measured by a comprehension and a transfer test, and strategy use, measured by self-report scales. Seventy-one students (Grade 10) participated in the study. The results showed that studying predefined summaries in a pictorial representation mode facilitated deep understanding. Furthermore, mediation analysis showed that the effect of representational mode was mediated by students' spatial representations of learning content. The effect of spatial representations was in turn facilitated by mental imagery activities.