Fiducial markers (FMs) are commonly used as a correction technique for interfraction translations of the prostate. The aim of this investigation was to determine the magnitude of prostate rotations using 2 methods: FM coordinates and the anatomical border of the prostate and rectum. Daily computed tomography (CT) scans (n = 346) of 10 prostate cancer patients with 3 implanted FMs were acquired using the CT on rails. FM coordinates were used to determine rotation in the sagittal, transverse, and coronal planes, and CT contours of the prostate and rectum were used to determine rotation along the sagittal plane. An adaptive technique based on a subset of images (n = 6; planning and first 5 treatment CTs) to reduce systematic rotation errors in the sagittal plane was tested. The standard deviation (SD) of systematic rotation from FM coordinates was 7.6°, 7.7°, and 5.0° in the sagittal, transverse and coronal planes. The corresponding SD of random error was 10.2°, 15.8°, and 6.5°. Errors in the sagittal plane, determined from prostate and rectal contours, were 10.1° (systematic) and 7.7° (random). These results did not correlate with rotation computed from FM coordinates (r = -0.017; p = 0.753, n = 337). The systematic error could be reduced by 43% to 5.6° when the mean prostate position was estimated from 6 CT scans. Prostate rotation is a significant source of error that appears to be more accurately determined using the anatomical border of the prostate and rectum rather than FMs, thus highlighting the utility of CT image guidance.