The objective of this study was to evaluate set-up uncertainties using a portal imaging system in a population of inoperable non-small cell lung cancer. Twenty-one patients were treated by a conformal radiotherapy technique with a personalized immobilization cast. The beam was verified by comparison with a corresponding digitally reconstructed radiograph by superimposition of anatomical structures. One thousand eight hundred eighty three images were analyzed. The mean intrafraction and interfraction errors (+/-SD) were 2.17 mm and 0.9 +/- 3.7 mm, 2.3 mm and 0.9 +/- 3.1 mm, 3 mm and 0.7 +/- 3 mm on the lateral (x), cranio-caudal (y) and anterior-posterior (z) axes, respectively. The mean systematic error was small, less than 1 mm, in all directions. The random errors were 2.5 mm, 2.4 mm, and 1.8 mm on the x, y, and z axes, respectively. No correlation between errors and the patient's height, weight, age, or sex was found. Set-up errors accuracy depending on practices, each institution should review their own treatments to quantify and reduce set-up errors in clinical practice.