The purpose of this study was to evaluate the uncertainty of prostate position relative to CT-planned position during proton therapy. Twenty-five patients enrolled in a clinical proton therapy study had 3 or 4 gold markers implanted into the prostate one week prior to radiation planning CT study. To minimize prostate motion, a water-filled endorectal balloon was placed; patients were instructed to comfortably fill the bladder by controlling water intake prior to CT and treatment. During the first five treatment days, orthogonal posterior-anterior and lateral x-ray films were taken immediately before and after treatment. The 3-D location of the centroid of the seeds was determined in reference to pelvic anatomical axes, defined during CT-based treatment planning. SDs of the prostate intrafractional motion about the daily mean and interfractional motion about the global mean position over five days were calculated. The mean and SD of the deviation of the global mean prostate position relative to the CT-planned position also were determined. From these SDs, the overall composite SD and a 99.7% confidence interval of the prostate position relative to the planned position were derived. The SD of intrafractional motion about the daily mean was 0.3 mm left-right (LR), 1.1 mm anterior-posterior (AP), and 1.1 mm inferior-superior (IS). The SDs of the interfractional motion were 0.4 mm LR, 1.6 mm AP, and 1.4 mm IS. The difference between the global mean and the CT-planned reference position for each patient was -0.2 ± 0.6 mm LR, 1.5 ± 3.2 mm AP, and -1.8 ± 2.4 mm (mean ± SD). Composite SDs for the uncertainty of prostate position relative to the CT-planned position were 0.8 mm LR, 3.8 mm AP, and 3.0 mm IS; the 99.7% confidence (3 SDs) margins were 2.4 mm LR, 11.3 mm AP, and 8.9 mm IS. Intra- and interfractional motion, and deviation of mean prostate position from CT reference position contribute to the uncertainty of prostate position with increasing magnitude. Required confidence margins including the prostate within field boundaries match current practice but can be reduced by using implanted gold markers.