Studies on using still images and dynamic videos in multimedia annotations produced inconclusive results. A further examination, however, showed that the principle of using videos to explain complex concepts was not observed in the previous studies. This study was intended to investigate whether videos, compared with pictures, better assist English learners to learn difficult words. It adopted a three-group immediate posttest and delayed posttest quasi-experimental design. Ten target words were selected and embedded in a reading text, each of which was annotated by three annotation types: text-only, text and picture, and text and video. Three intact classes, a total of 88 students, were recruited in a junior high school in northern Taiwan, each of which was randomly assigned to one of the three groups. All participants took the pretest two weeks before the experiment, the immediate posttest after reading the text, and the delayed posttest two weeks after the experiment. The result revealed significant differences between the three groups, in which the video group outperformed the other two groups. Pedagogical implications and suggestions for future research are also given. (Contains 5 tables and 1 figure.)