Abstract
Knowing when one has understood is an important metacognitive skill in reading and in making use of what one has read. It is defined here as the ability to detect failures of comprehension processing. This is measured operationally, using a signal detection model, as the extent to which a respondent's judgments of confidence discriminate between accurate and inaccurate responses on cloze tests. The reliability of the measure is investigated and, as an illustration, it is applied to reading in native and second languages. It is concluded that this new augmented form of cloze testing may be a useful tool in reading research.
