This paper describes a four-step approach to constructing a diagnostic test that provides precise but practical information on students' problems and needs for additional instruction or remediation. The approach is based on analyzing the structure of the domain to determine which skills within the domain need to be assessed to diagnose students' problems. The four steps include: (1) identifying the factors that describe the curricular domain, (2) constructing a test with items representing all possible combinations of content and cognitive factors, (3) determining which factors and interactions among them produced variations in students' scores using generalizability theory, and (4) determining the minimum number of items needed to obtain a generalizable measure of each skill in the diagnostic profile. This paper contains three studies: "Optimizing the Diagnostic Power of Tests: An Illustration from Language Arts," by Noreen Webb, Joan Herman, and Beverly Cabello; "Diagnosing Student Errors: An Example from Science," by Steve Shaha; and, "Task Structure Design: Beyond Linkage," by Eva Baker and Joan Herman. Two of these studies explicity address problems related to the design of diagnostic tests; the third considers a conceptual model for integrating testing and instruction. (PN)