A study explored the use of a text-copying task for the determination of syntactic constituent structure. It was predicted that the task would be a naturalistic, reasonably direct, and sensitive psycholinguistic research method. In two experiments, 70 subjects wrote out copies of typed passages. The points where they paused and looked back to the typed originals were recorded and aggregated across all subjects copying a given passage. The passages differed in content but contained sentences with comparable structure. In general, patterns of pausing were highly correlated across sentences that matched in structure, but reliably distinct in relevant control comparisons. Sentences, clauses, and many details of their internal constituency emerge clearly. These details are presented and compared with descriptions from the field of theoretical linguistics. Some suggestive evidence in support of two-stage parsing models is noted. (MSE)