The purpose of the present study was to compare the impact of pulmonary function, body habitus, and stereotactic body radiation therapy (SBRT) immobilization on setup and reproducibility for upper lung tumor. From 2008 through 2011, our institution's prospective SBRT database was searched for patients with upper lung tumors. Two SBRT immobilization strategies were used: full-length BodyFIX and thermoplastic S-frame. At simulation, free-breathing, four-dimensional computed tomography was performed. For each treatment, patients were set up to isocenter with in-room lasers and skin tattoos. Shifts from initial and subsequent couch positions with cone-beam computed tomography (CBCT) were analyzed. Accounting for setup uncertainties, institutional tolerance of CBCT-based shifts for treatment was 2, 2, and 4 mm in left-right, anterior-posterior, and cranial-caudal directions, respectively; shifts exceeding these limits required reimaging. Each patient's pretreatment pulmonary function test was recorded. A multistep, multivariate linear regression model was performed to elucidate intervariable dependency for three-dimensional calculated couch shift parameters. BodyFIX was applied to 76 tumors and S-frame to 17 tumors. Of these tumors, 41 were non-small cell lung cancer and 15 were metastatic from other sites. Lesions measured < 1 (15%), 1.1 to 2 (50%), 2.1 to 3 (25%), and > 3 (11%) cm. Errors from first shifts of first fractions were significantly less with S-frame than BodyFIX (p < 0.001). No difference in local control (LC) was found between S-frame and BodyFIX (p = 0.35); two-year LC rate was 94%. Multivariate modeling confirmed that the ratio of forced expiratory volume in the first second of expiration to forced vital capacity, body habitus, and the immobilization device significantly impacted couch shift errors. For upper lung tumors, initial setup was more consistent with S-frame than BodyFIX, resulting in fewer CBCT scans. Patients with obese habitus and poor lung function had more SBRT setup uncertainty; however, outcome and probability for LC remained excellent.