Traditional cloze tests (such as the CBM-maze) may be poor measures of comprehension processes beyond the single sentence level. This paper presents an alternative, a deep cloze test with gaps that are strategically chosen to assess comprehension beyond the sentence level. To fill each gap, the reader has to draw global cohesion inferences during reading. A study with 83 adult students found support for the validity of the deep cloze test. It contributed unique variance to global reading comprehension as measured with a conventional test even after controlling for sentence-level cloze, word decoding, and vocabulary knowledge. The deep cloze also explained all of the shared variance between reading comprehension and a topic identification task designed to require global comprehension. Future studies may explore how the deep cloze format can be used to tap other component processes of reading comprehension.