This paper describes a novel method to create a quantitative model of an educational content domain of related practice item-types using learning curves. By using a pairwise test to search for the relationships between learning curves for these item-types, we show how the test results in a set of pairwise transfer relationships that can be expressed in a Q-matrix domain model. Creating these Q-matrices for various test criteria we show that the new domain model results in consistently better learning curve fits as shown by cross-validation. Further, the Q-matrices produced can be used by educators or curriculum designers to gain a richer, more integrated perspective on concepts in the domain. The model may also have implications for tracing student knowledge more effectively to sequence practice in tutoring/training software. (Contains 2 figures.) [This research was funded by the U.S. Department of Education, Carnegie Learning Inc., the Pittsburgh Science of Learning Center and DataShop team, and Ronald Zdrojkowski.]