One common way of assessing learners' proficiency in second language (L2) listening comprehension is by comprehension-restitution (C-R), whereby learners write down what they have heard. The various forms of C-R differ according to characteristics of the input (e.g., length, speech rate) and of the output (e.g., amount of detail requested). The procedure may thus make different demands in terms of how the input is processed and written down. Analysis of the C-R procedure raises issues both about the assessment and teaching of listening comprehension and about the processes involved in the accomplishment of the task. The performance of eight French learners of English was examined, and discussed in interviews, with regard to three C-R tasks: paragraph restitution, sentence completion, and summarising. The criteria adopted were word recognition, intelligibility of answers, and task accomplishment. Results show that a high percentage of words needs to be recognised before the task can be accomplished, but that top-down processes can help in reaching an intelligible answer. Implications for teaching are discussed. (Contains 3 tables and 1 footnote.)