In this quasi-experimental study, 165 learners of English for academic purposes at a university in China were randomly assigned to five experimental groups and a control group. Each experimental group encountered 19 target collocations in the same academic lecture in one of the following input modes: (a) reading, (b) listening, (c) reading while listening, (d) viewing, and (e) viewing with captions. The control group did not receive any treatment. The results revealed that reading, viewing, and viewing with captions led to learning at the form recognition level, but no significant differences were found in the learning gains across these modes. Nonverbal elaboration, type of vocabulary, and type of verbal elaboration affected learning, but frequency of occurrence, strength of association, comprehension, and prior knowledge of general vocabulary did not. This study provides further evidence supporting the use of academic lectures for incidental learning of collocations as well as expanding on the multimedia learning theory.