This report is based on the notion that criteria for "successful" diagnosis differ for different medical specialties. Diagnostic judgement in highly unstructured task domains is best characterized as clinical problem formulation with pragmatic implications for interventions. This concept is illustrated by the implementation of a computer-based diagnostic expert system for geriatric psychiatry. The program begins with guiding the patient's interview by the clinician. This is followed by the diagnostic process which is conducted in three steps. In the first step, candidate syndromes of disease states are identified by matching the presenting complaints with symptom lists of clinical syndromes. In the second step, the candidate syndromes serve as the organizers for gathering additional patient data (e.g., information from referral source and past illnesses) to seek constellations of findings which form a coherent clinical context. In this phase, the program can also prompt the clinician to obtain additional information from the patient or other available sources to resolve conflicts or, if necessary, invoke the concurrent presence of two independent syndromes to attain convergence (i.e., minimize the number of unexplained findings). In a final step, the syndromes with highest credence value, reflected by the number of linkages to patient data, are sharpened to specific diagnostic formulations. This is accomplished by frames of diagnoses which are associated with each syndrome. Once again, the patient data can be consulted to achieve optimal matches. A summary statement to the clinician lists the essential supporting evidence for the chosen diagnosis and can communicate alternatives or unresolved issues for clarifying interventions. Although no decisive evidence is available, informal observations of clinicians support the notion that clinical reasoning in highly unstructured domains is largely contextual and seeks apprehension of "gestalts" in successive steps. The diagnostic program of this report aims at simulating this process.