One of the challenges of an intelligent tutoring system (ITS) is adapting the difficulty level of the questions posed to the student to suit the student’s academic level. Our study examines the task of adjusting the system’s level of challenges to the level of the learner and addresses the questions of how best to do so and whether there is any benefit from such adjustment. To answer these questions, we developed reading comprehension courseware that includes three adaptive algorithms for adjusting the level of the questions presented to the students: the random selection algorithm, the Q-learning based algorithm, and the Bayesian inference algorithm. We conduct a real-world experiment in which real high school students used the courseware to improve their reading comprehension skills. In order to compare and evaluate the performance of the algorithms, the courseware used by each student utilized one of the three adaptive algorithm alternatives. Our results demonstrate that when considering all of the students, there was significant improvement (learning gain) using each of the methods.