Three studies were conducted with middle school students to evaluate a web-based intelligent tutoring system (ITS) for arithmetic and fractions. The studies involved pre and post test comparisons, as well as group comparisons to assess the impact of the ITS on students' math problem solving. Results indicated that students improved from pre to post test after working with the ITS, whereas students who simply repeated the tests showed no improvement. Students who had more sessions with the ITS improved more than those with less access to the software. Improvement was greatest for students with the weakest initial math skills, who were also most likely to use the multimedia help resources for learning that were integrated into the software. (Contains 4 tables.)