To determine the residual setup errors of several image guidance scenarios, using cone-beam computed tomography (CBCT) in conventionally fractionated radiotherapy for lung tumors. Thirteen lung cancer patients were treated with conventionally fractionated radiotherapy, using daily image guidance with CBCT, resulting in 389 CBCT scans which were registered to the planning scan using automated soft-tissue registration. Using the resulting daily alignment data, 4 imaging frequency scenarios were analyzed: (A) no imaging; (B) weekly imaging with a 3-mm threshold; (C) first 5 fractions imaged, then weekly imaging with a patient-specific threshold; and (D) imaging every other day. The systematic setup error (Sigma) was reduced with increasing frequency of imaging from 3.4 mm for no imaging to 1.0 mm for imaging every other day. Random setup error (sigma), however, varied little regardless of the frequency of imaging: 2.9, 3.0, 3.4, and 3.2 mm for scenarios A, B, C, and D, respectively. The setup margins required to account for the residual error of each imaging scenario were 1 to 1.6 cm for scenario A, 4 to 6 mm for scenarios B and C, and 4 to 5 mm for scenario D. As the residual error of daily CBCT was not included in this analysis, these margins compare with a margin of zero for daily CBCT. Daily image guidance is ideal as the setup margin can be reduced by about 5 mm versus a nondaily imaging scenario. However, if daily image guidance is not possible, there is little benefit in imaging more often than once a week.