Abstract Despite the intuitively compelling adage ‘a picture is worth a thousand words’, a attempts over the past decade to use animations to explain algorithms to students have produced disappointing results. In most cases, interesting algorithm animations were designed, but no formal, systematic evaluations were conducted. When such evaluations were performed the results were mixed, with compelling evidence for the instructional superiority of algorithm animations failing to emerge. It is in this context that we embarked on a research program to develop educationally effective algorithm visualizations. This program was based on the premise that animations needed to be embedded in a knowledge and context providing hypermedia environment in order to effectively harness their power to enhance learning. This paper describes the architecture of the resulting Hypermedia Algorithm Visualization system (HalVis). Four empirical studies with HalVis are described, which demonstrated that the extent of learning exhibited by students who used HalVis was significantly greater than that of students who used means of traditional instruction or a typical algorithm animation.