Cancer patients often experience multiple symptoms concurrently, a phenomenon called symptom clustering. Different symptom clusters have been identified by various symptom assessment tools, as well as by different research methods, but no study has reported whether these identified symptom clusters can be replicated in a new sample. The severity of nine symptoms in 321 cancer patients was assessed using a Taiwanese version of the M.D. Anderson Symptom Inventory. The fit between these data and a model with three symptom factors (sickness, gastrointestinal, and emotional) was evaluated using confirmatory factor analysis. Most fitness indices demonstrated a satisfactory fit between the data and a prespecified three-factor model except one; the root mean square error of approximation was <0.06. A modified model with one symptom (lack of appetite) double loaded in the sickness and gastrointestinal factors demonstrated a significantly better fit between the data and the model. Higher scores in each of the three symptom factors were associated with poorer functional status. Metastatic disease and receiving both chemotherapy and radiation therapy were associated with higher scores in sickness and gastrointestinal factors, but not in the emotional factor. Only hospitalization affected patients' scores in emotional factors. Our findings confirmed the prespecified structure of symptom clusters. A modified model showed a better fit. Patients' complex symptom experience may be better represented by subscale scores based on meaningful clusters rather than on an overall score across all symptoms.