A three year longitudinal intervention study is presented where two conditions of reading instruction are compared. In a base condition six teachers--four in the experimental groups and two in the control groups--and their students in grade 4-6 discussed expository texts during a regular lesson. The discussions, which took place in small groups, six students in each group, were videotaped. The majority of the students were poor comprehenders. In the intervention condition, the teachers in the experimental groups used a new approach of comprehension instruction, Questioning the Author, after repeatedly having participated in seminars led by the investigator where the model was practised. Dependent variables were the type of questions asked by the teacher, and the students' response behaviour and interaction with the texts. During the regular lessons, most of the teachers asked purely factual questions where the students just had to retrieve information from the text to be able to answer them. The students made few inferences and reflections. During Questioning the Author lessons in the experimental groups, the teachers' question types had undergone a change. The number of purely factual questions had decreased. Instead there was an increase in the number of inferences and half-open questions. This was not the case in the control groups. The students interacted more actively with the text during Questioning the Author lessons. They clearly made an effort to explore ideas in the text and made numerous inferences and reflections. The students in the control groups seldom got the opportunity to wrestle with text ideas.