To investigate the causes of variation in cloze-comprehension test correlations, a reanalysis was conducted of the influential J.R. Bormuth study (1962), which reported correlations between nine cloze and nine comprehension tests administered to 50 subjects in grades 4, 5, and 6. Two separate renalayses were completed in the present study, the first using cloze-comprehension correlations as a dependent measure and the second using correlations of cloze with each of the Bormuth question catagories--vocabulary, factual recall, sequential order, cause and effect, inference, and author's purpose--at each grade level. Results indicated that in both reanalyses, variable reliability and grade level differences accounted for some of the variation in correlations. The proportion of comprehension test questions found to assess within-sentence information was found to be significant in the first regression analysis but not in the second. Results suggested that the limited construct validity of cloze as a measure of comprehension had a small but significant influence on the concurrent validity of cloze with measures of comprehension. Cloze predicted performance best when within-sentence comprehension dominated, but was less useful when students were expected to integrate information across sentence boundaries. (MM)