We compared performances on a learning task in which students (N = 81) viewed a pedagogical multimedia document without (control group) or with a readymade graphic organizer (readymade group) with performances on an active learning task where students self-generated a graphic organizer either totally (total self-generated group) or partially (partial self-generated group) while learning from the same multimedia document. According to the "generative" hypothesis, asking students to actively engage in the construction of a graphic organizer enhances their learning, owing to the generative processes (selection, organization, integration) required to perform the task. However, according to the "cognitive load" hypothesis, generating a graphic organizer can hinder students' learning, owing to the extraneous processing elicited by the task. It can nonetheless be assumed that if scaffolding is provided to students in the shape of an empty graphic organizer to fill in, these negative effects can be avoided. Results confirmed the beneficial effect of providing a graphic organizer on students' retention of the elements contained in the multimedia document (macrostructure information, hierarchical relations). Evidence in favor of the cognitive load hypothesis and against the generative hypothesis was found, as students in the total self-generated group performed more poorly on the retention and transfer tests than those in the readymade group. This negative effect on learning ceased to be observed when scaffolding was provided to students in the partial self-generated group, although they still spent more time on the document than those in the readymade group. Overall, we failed to observe any beneficial effect of generation on learning.