The aim of this study is to verify a lessonware effect concerning the comprehension level of 8- to 11-year-old children as they improve their comprehension monitoring (Baker, 1984). Kintsch (1998) argues that comprehension contains three levels of information-representation: microstructure, macrostructure and a situational model. Microstructure is low level comprehension used to comprehend a text, proposition by proposition. Macrostructure organizes several propositions from the microstructure to create an ideal text summary. The situational model is a high level representation, which assembles text information and reader's knowledge in order to construct a mental representation of the text. Each comprehension level requires coherence processing to verify that the construction is consistent. This coherence processing, also called comprehension monitoring, detects lexical, syntactic, and semantic inconsistencies and attempts to resolve these by way of inference (Baker, 1985). Baker divides the semantic criterion into five categories: two on propositional cohesiveness (propositions must be compatible with the text thematic and linked with shared arguments), two on propositional coherence (checking that a proposition in the text is not inconsistent with another proposition or with domain knowledge) and one on global text organization. Oakhill, Cain & Yuill (1996) show that poor comprehension monitoring is directly linked to low- achievers' reading comprehension skills. Many studies describe monitoring development but few attempt to teach monitoring. Kinnunen & Vauras (1995) succeeded in improving the level of comprehension of low-achievers by training nine-year-old children on lexical and coherences criteria. We assume this training may be more efficient if it were better differentiated. Monitoring studies (Baker, 1985) show high individual variation on monitoring acquisition. Kinnunen & Vauras studied only three sub-criteria of semantic and global performance. We studied four monitoring criteria (lexical, two on coherences and cohesions) and test the comprehension level of each of them.Pre-testing was carried out by computer to obtain monitoring and comprehension level of 500 children (half of these were assigned to a control group). Each child read thirty stories including six texts of each lexical, external, internal and anaphoric inconsistency. Each story was displayed line by line on a screen's computer in a self-paced manner. We consider that a criterion was satisfied when a child's reading time of a target line (comprising an inconsistency) is significantly higher than his/her mean reading time for consistent stories. Four questions (testing all levels of comprehension) followed each story. Pre-testing allowed us to form three levels based on the number of criteria monitored by children: less than two, two or three, and more than three. Crossley & Green's (1990) method was used to construct our lessonware, which is composed of a lesson, a demonstration, and guided training (teachers helped us construct the contents of the lessonware). This method was coupled with a direct explanation approach which recommends an explanation of children's responses by the teacher. Guided training allows a child to be trained with respect to inconsistency detection. Four didactical sequences were created for each criterion investigated (and one mixed all four criteria). During guided training, children were required to answer questions. If a response was false, a demonstration describes how an expert would respond to the exercise (and how an expert would detect the inconsistency in question). Each child's coursework was automatically saved after each session. A post-test (which were similar to the pre-test) will allow us to observe and compare children's final level of comprehension-monitoring. Post- test results and analyses (which are underway) will be compared to pre-test results.