BACKGROUND: During chemotherapy, some colorectal liver metastases (LMs) disappear on serial imaging. This disappearance may represent a complete response (CR) or a reduction in the sensitivity of imaging during chemotherapy. The objective of the current study was to determine the fate of disappearing LMs (DLMs) and the factors that predict a true CR. METHODS: Between 2000 and 2003, 435 patients who were evaluated by hepatobiliary surgeons received chemotherapy before they were considered for resection. Inclusion criteria were <12 LMs before chemotherapy, at least 1 DLM on a computed tomography (CT) scan, and either surgical resection or 1 year of clinical follow-up after the disappearance of LMs. A true CR was defined as either a pathologic CR (no tumor detected in the resection specimen) or a durable clinical CR (did not reappear on follow-up imaging). Clinical and pathologic factors were analyzed to identify those associated with a true CR. RESULTS: During chemotherapy, 39 patients (9%) had a total of 118 DLMs on follow-up CT scans. Sixty-eight DLMs were resected, and 50 were followed clinically. Overall, 75 DLMs (64%) were true CRs, including 44 pathologic CRs and 31 durable clinical CRs. On multivariate analysis, the use of hepatic arterial infusion (HAI) chemotherapy (odds ratio [OR], 6.2; P ¼ .02), the inability to observe the DLM on a magnetic resonance image (OR, 4.7; P ¼ .005), and normalization of serum carcinoembryonic antigen levels (OR, 4.6; P ¼ .006) were associated independently with a true CR. CONCLUSIONS: Approximately 66% of DLMs represented a true CR according to assessment by resection or radiologic follow-up. Predictive factors may help to stratify patients who are likely to harbor residual disease. Cancer 2010;116:1502-9.