The purpose of this study was to identify independent predictors of clinically significant fatigue based upon a multidimensional model. A total of 180 cancer patients completed the Brief Fatigue Inventory (BFI), Functional Assessment of Cancer Therapy-Fatigue (FACT-F), Memorial Symptom Assessment Scale Short Form (MSAS-SF), and the Zung Self-Rating Depression Scale (SDS). Additional data included Karnofsky Performance Status (KPS) score, laboratory tests, and demographic information. The BFI usual fatigue severity > or =3/10 was defined as clinically significant fatigue. Possible independent variables were identified from a biopsychosocial model of fatigue. Fisher's exact test was used to univariately assess the association of each variable with clinically significant fatigue. Multiple logistic regression analyses were used to identify independent predictors of fatigue within each dimension, and then across dimensions. Fatigue was present in 113 (62%) patients, and 80 (44.4%) patients had usual fatigue > or =3/10. The unidimensional independent predictors were use of analgesics (situation dimension); hemoglobin and serum sodium (biomedical dimension); feeling drowsy, dyspnea, pain and lack of appetite (physical symptom dimension); and feeling sad and feeling irritable (psychological symptom dimension). In a multidimensional model, dyspnea, pain, lack of appetite, feeling drowsy, feeling sad, and feeling irritable predicted fatigue independently with good calibration (Hosmer Lemeshow Chi Square=5.73, P=0.68) and discrimination (area under the receiver operating characteristic curve=0.88). Physical and psychological symptoms predict fatigue independently in the multidimensional model, and superseded laboratory data. These findings support a symptom-oriented approach to assessment of cancer-related fatigue.