In the present study, we examined the use of verbal protocols as data in the study of the cognitive processes underlying insight. Fifty-eight Temple University undergraduates attempted to solve Duncker’s (1945) candle problem either silently or while thinking aloud. Solution rates, solving times, and solution types were comparable between conditions, suggesting thatverbal overshadowing (Schooler, Ohlsson, & Brooks, 1993) did not occur when the participants attempted to solve the candle problem. Subsequent analysis of verbal protocols provided a catalogue of solutions generated by the participants, as well as empirical support for the occurrence of impasse and restructuring. Although restructuring was present in the majority of protocols, including those of the participants who later produced the box solution, the presence of impasse occurred with less frequency and was not associated with production of the box solution. These results provide information concerning how the candle problem is solved and suggest that verbalization can be used to examine how individuals solve insight problems and to evaluate existing theories of insight.