The widespread application of computer technologies to language learning has led to increasingly academic reading taking place in a CALL environment. This provides opportunities for multimedia annotations to assist the reading process. However, despite the increased interests and emerging literature in multimedia annotations, research has revealed wide variation in their effectiveness for reading comprehension and insufficient investigation into the mediating role of individual difference factors in annotation use. This study focuses on the effect of annotation use on L2 reading comprehension. In particular, it explores whether L2 reading is influenced by different types of annotations, and whether working memory (WM) mediates annotation use in influencing reading performance. One hundred and sixty-two ESL learners from a Chinese university participated in the study. They were identified to be low, medium or high in WM capacity based on the composite z-scores of three WM measures. The study required all participants to read four texts of the same difficulty, each text under a different condition: no annotations, verbal annotations, visual annotations, or verbal plus visual annotations. These conditions were counterbalanced across the readings. Participants completed a reading comprehension test for each reading, and a follow-up questionnaire on their perceptions of annotations. Fifteen participants were further interviewed to provide supplementary qualitative data concerning their reading strategies, interactions with annotations, and views of annotation design. The findings were mainly discussed with reference to the interactive model of reading, the theory of WM, the cognitive theory of multimedia learning, and the qualitative data from questionnaire and interview surveys. Results revealed that (a) annotation use led to improved reading comprehension, (b) the combined annotation treatment was most beneficial to reading comprehension, (c) annotation use benefited the low WM learners the most and the high WM learners the least, and (d) annotation types did not make a difference to reading comprehension for learners of similar WM capacity. The results thus suggest that annotations can enhance reading comprehension to varying degrees, and support the crucial role of WM in reading comprehension.