To investigate the clinical importance and feasibility of a 3-D portal image analysis method in comparison with a standard 2-D portal image analysis method for pelvic irradiation techniques. In this study, images of 30 patients who were treated for prostate cancer were used. A total of 837 imaged fields were analyzed by a single technologist, using automatic 2-D and 3-D techniques independently. Standard deviations (SDs) of the random, systematic, and overall variations, and the overall mean were calculated for the resulting data sets (2-D and 3-D), in the three principal directions (left-right [L-R], cranial-caudal [C-C], anterior-posterior [A-P]). The 3-D analysis included rotations as well. For the translational differences between the three data sets, the overall SD and overall mean were computed. The influence of out-of-plane rotations on the 2-D registration accuracy was determined by analyzing the difference between the 2-D and 3-D translation data as function of rotations. To assess the reliability of the 2-D and 3-D methods, the number of times the automatic match was manually adjusted was counted. Finally, an estimate of the workload was made. The SDs of the random and systematic components of the rotations around the three orthogonal axes were 1. 1 (L-R), 0.6 (C-C), 0.5 (A-P) and 0.9 (L-R), 0.6 (C-C), 0.8 (A-P) degrees, respectively. The overall mean rotation around the L-R axis was 0.7 degrees, which deviated significantly from zero. Translational setup errors were comparable for 2-D and 3-D analysis (ranging from 1.4 to 2.2 mm SD and from 1.5 to 2.5 mm SD, respectively). The variation of the difference between the 2-D and 3-D translation data increased from 1.1 mm (SD) for zero rotations to 2.7 mm (SD) for out-of-plane rotations of 3 degrees, due to a reduced 2-D registration accuracy for large rotations. The number of times the analysis was not considered acceptable and was manually adjusted was 44% for the 2-D analysis, and 6% for the 3-D analysis. True 3-D analysis of setup errors for a group of 30 patients with prostate cancer demonstrated that setup rotations are rather small. The deformation of the projected anatomy in portal images caused by out-of-plane rotations leads to a reduced 2-D registration accuracy. For rotations larger than 3 degrees this effect can be quite pronounced, making 3-D registration the preferred method. Furthermore, the automatic 3-D registration has a higher success rate, most likely because this technique uses more information compared to the 2-D method.