This study was performed as part of a large depression screening project in cancer patients to determine the degree of physician recognition of levels of depressive symptoms in cancer patients and to describe patient characteristics that influence the accuracy of physician perception of depressive symptoms. Twenty-five ambulatory oncology clinics affiliated with Community Cancer Care, Inc of Indiana enrolled and surveyed 1,109 subjects treated by 12 oncologists. Subjects completed the Zung Self-Rating Depression Scale (ZSDS) and physicians were asked to rate their patients' level of depressive symptoms, anxiety, and pain using numerical rating scales. Subjects' sex, age, primary tumor type, medications, primary caregiver, and disease stage at diagnosis were also recorded. Physician ratings of depression were significantly associated with their patients' levels of endorsement of depressive symptoms on the ZSDS. However, agreement between physicians and patients is most frequently clustered when patients report little or no depressive symptoms. While physician ratings are concordant with patient endorsement of no significant depressive symptomatology 79% of the time, they are only concordant 33% and 13% of the time in the mild-to-moderate/severe ranges, respectively. Physician ratings were most influenced by patient endorsement of frequent and obvious mood symptoms, ie, sadness, crying, and irritability. Physician ratings also appeared to be influenced by medical correlates of patients' level of depressive symptoms (functional status, stage of disease, and site of tumor). Additionally, patients whose depression was inaccurately classified reported significantly higher levels of pain and had higher levels of disability. Physicians' ratings of depression were most highly correlated with physicians' ratings of patients' anxiety and pain. Physicians' perceptions of depressive symptoms in their patients are correlated with patient's ratings, but there is a marked tendency to underestimate the level of depressive symptoms in patients who are more depressed. They are most influenced by symptoms such as crying and depressed mood, and medical factors that are useful, but not the most reliable, indicators of depression in this population. Physicians' ratings of their patients' distress symptoms seem to be global in nature--they are highly correlated with anxiety, pain, and global dysfunction. Physician assessment might be improved if they were instructed to assess and probe for the more reliable cognitive symptoms such as anhedonia, guilt, suicidal thinking, and hopelessness. Screening instruments and the use of brief follow-up interviews would help to identify patients who are depressed.