The construct validity of measures of reading comprehension was discussed. A word deletion procedure was proposed consistent with a model of comprehension which emphasizes the importance of context. The procedure assures sensitivity to context by allowing only highly constrained words to be deleted and by increasing the information value of context. Third and sixth grade students answered word deletion items at three levels of contextual constraint. The results indicated that while sixth graders were more skilled than third graders, both groups could use contextual information to reduce uncertainty. The group differences were not observed for items which were not highly constrained. That highly constrained items were more difficult and yielded a more narrow range of responses than less constrained items provided evidence of the construct validity of the proposed measure. (Author)