PURPOSE: Kilovoltage cone-beam computed tomography (CBCT) has been developed to provide accurate soft-tissue and bony setup information. We evaluated clinical CBCT setup data and compared CBCT measurements with electronic portal imaging device (EPID) images for lung cancer patients. METHODS AND MATERIALS: The setup error for CBCT scans at the treatment unit relative to the planning CT was measured for 62 patients (524 scans). For 19 of these patients (172 scans) portal images were also made. The mean, systematic setup error (Sigma), and random setup error (sigma) were calculated for the CBCT and the EPID. The differences between CBCT and EPID and the rotational setup error derived from the CBCT were also evaluated. An offline shrinking action level correction protocol, based on the CBCT measurements, was used to reduce systematic setup errors and the impact of this protocol was evaluated. RESULTS: The CBCT setup errors were significantly larger than the EPID setup errors for the cranial-caudal and anterior-posterior directions (p <0.05). The mean overall setup errors after correction measured with the CBCT were 0.2 mm (Sigma = 1.6 mm, sigma = 2.9 mm) in the left-right, -0.8 mm (Sigma = 1.7 mm, sigma = 4.0 mm) in cranial-caudal and 0.0 mm (Sigma = 1.5 mm, sigma = 2.0 mm) in the anterior-posterior direction. Using our correction protocol only 2 patients had mean setup errors larger than 5 mm, without this correction protocol 51% of the patients would have had a setup error larger than 5 mm. CONCLUSION: Use of CBCT scans provided more accurate information concerning the setup of lung cancer patients than did portal imaging.