Although many talented researchers have created excellent tools for computer-assisted instruction and intelligent tutoring systems, creating high-quality, effective, scalable but individualized tools for learning at a low cost is still an open research challenge. Many learning tools create complex models of student behavior that require extensive time on the part of subject experts, as well as cognitive science researchers, to create effective help and feedback strategies. In this research, we propose a different approach, called the q-matrix method, where data from student behavior is “mined” to create concept models of the material being taught. These models are then used to both understand student behavior and direct learning paths for future students. We describe the q-matrix method and present preliminary results that imply that the method can effectively predict which concepts need further review.