A study was conducted to investigate the degree to whicb subjects could be trained to improve the reliability of their diagnostic decisions about cases of reading difficulties. Twenty-eight graduate students enrolled in a reading diagnosis course were trained from G. B. Sherman's Model of Reading and Learning to Read, and practiced on simulated cases of reading difficulty. Pretests and posttests were administered, in which the students were asked to diagnose cases, write their diagnoses, and transfer their diagnoses to checklists. Statistics were run to calculate the students' diagnostic agreement with each other. The students' diagnostic agreement approximately doubled from pretests to posttests, indicating that subjects can be trained to improve the reliability of their diagnostic decisions about cases of reading difficulties. (Eleven appendixes include copies of materials used in the course.) (Author/MKM)