This research investigates comprehension monitoring by studying the relation between the perception of comprehension during reading and detection of comprehension failures. Two groups of four college students (good and poor learners) defined by a post-hoc median split learned a physics chapter while rating their comprehension on a four-point scale and justifying it after each sentence. It was found that poor learners process sentences at the local level and use fuzzy referents to talk about the text, while good learners are more exact, specific, and global. Poor learners detect many of their comprehension failures, but they pay more attention to what they understand than to what they do not understand, which leads them to label their comprehension as "approximate" instead of"poor" even for severe comprehension failures and to go on reading instead of trying to solve their problems. Poor learners are also deficient at evaluating hypotheses that they make to solve comprehension failures. Thus, the proposed new paradigm shows that detection of a comprehension problem does not necessarily mean good comprehension monitoring as is typically the case in the traditional error detection paradigm. Three tables present study data and there is a 32-item list of references. (Author/SLD)