To investigate the properties of a mathematical model for probability learning, response protocols were simulated to fit the models specifications, and then they were analyzed by the generalized learning curves technique. The results indicated that the generalized learning curves technique could discriminate among groups on a wide variety of bases. It was also found that the use of different event protocols for each subject is undesirable when trying to identify functional relations in the data. A procedure is proposed for the use of generalized learning curves technique in the development and refinement of mathematical models of learning.