The current study examined the effects of a computerized training program on reading comprehension, confidence ratings, and calibration of comprehension in adolescents with poor and good reading comprehension. Ninety 10th graders participated in the study and completed three training sessions. In each session, participants read two expository texts and answered multiple-choice questions. For each answer they gave, participants also rated their confidence. Participants were assigned to one of three online training conditions that differed in the type of immediate feedback provided after each question: (a) "Feedback on performance"; (b) "Feedback on performance and on calibration"; (c) "Feedback on performance with scaffolding (a cue for correcting wrong answers)". Results demonstrated that scaffolding feedback was the most effective training condition, leading to improved comprehension performance and calibration, especially for poor comprehenders. These findings highlight the necessity of developing theoretical and practice models of online feedback interventions for reading comprehension and self-evaluation abilities.