Two issues that may influence the diagnosis of depression in the medically ill are 1) the severity with which symptoms must be expressed before they are considered clinically significant and 2) how to deal with somatic symptoms that may be caused by medical illness. This study used different approaches to case identification to examine prevalence rates for major and minor depression in a group of terminally ill cancer patients. Semistructured diagnostic interviews were conducted with 130 patients receiving palliative care. Diagnoses according to the Research Diagnostic Criteria (RDC) were compared with diagnoses according to Endicott's revised criteria (which involve replacing somatic symptoms with non-somatic alternatives) when either a low-severity or a high-severity threshold for classifying RDC criterion A symptoms was used. A low-threshold (less stringent) diagnostic approach greatly increased the overall prevalence of major and minor depressive episodes with both the RDC and the Endicott criteria. With high thresholds, the RDC and the Endicott criteria were equivalent, whereas with low thresholds the Endicott substitutions identified fewer cases of major (but not minor) depression. Small differences between investigators in the applications of symptom-severity thresholds can result in large differences in prevalence rates for depression. However, the inclusions of somatic symptoms in the diagnostic criteria inflates the rates of diagnosis only when these symptoms are used in conjunction with a low-threshold approach.