The present study investigates how skilled and less skilled readers answered a gap-filling (or rational deletion cloze) test in order to examine whether such tests can measure text-level processing ability. Twelve Japanese EFL students (six skilled and six less skilled readers) completed a gap-filling test while thinking aloud about their test-taking processes. Their verbal protocols were categorized according to the classification of cloze item types developed by Bachman (1985). Results showed that both skilled and less skilled readers used text-level information more frequently than other types of information. The skilled readers, however, used text-level information more frequently than the less skilled readers. Qualitative analysis of the protocols revealed further differences in the use of different types of information between the groups. Although there were several cases where the items were answered correctly with local grammatical clues and extra-textual background knowledge, overall, the gap-filling test generated text-level processing and differentiated well between skilled and less skilled readers. The present study, therefore, supports the claim that a gap-filling test can be used as a test to measure higher order processing ability.