Abstract After having been out of fashion for some time, verbal reports have once again become established by certain authors as indicators of cognitive processes. In this perspective we have done an evaluation of different modalities of verbalization all related to instructions of a “thinking aloud” kind for a sorting task: (a) simultaneous verbalization (the subject is asked to verbalize what he says to himself while performing the task); and (b) subsequent verbalization (he is asked to verbalize what he said to himself at first without recall aids, afterwards in front of the record of his anterior behavior). The subjects performed three trials so that their activity becomes more-or-less routine. We have dealt with: (1) the compatibility between verbalization and the activity it refers to; and (2) the kind of verbal reports produced. It is shown that (a) simultaneous verbalization slows down the automation of the activity, introducés hitches and must be avoided out of problem-solving situations, (b) aided subsequent verbalization permits the automation, produces satisfactory and very precise verbal reports, as simultaneous verbalization does, and (c) unaided subsequent verbalization produces too much distance from the task and not very valid data. We conclude on the interest of aided subsequent verbalization.