The online image-guided localization data from 696 ultrasound (US), 598 mega-voltage cone-beam computed tomography (MV-CBCT), and 393 seed markers (SMs) couch alignments for patients undergoing intensity modulation radiotherapy of the prostate were analyzed. Daily US, MV-CBCT and SM images were acquired for 19, 17 and 12 patients, respectively, after each patient was immobilized in a vacuum cradle and setup to skin markers as the center of mass. The couch shifts applied in the lateral (left-right/LR), vertical (anterior-posterior/AP), and longitudinal (superior-inferior/SI) directions, along with the magnitude of the three-dimensional (3D) shift vector, were analyzed and compared for all three methods. The percentage of shifts larger than in all directions was also compared. Clinical target volume-planning target volume (CTV-to-PTV) expansion margins were estimated based on the localization data with US, CB, and SM image guidance. Results show the US data have greater variability. Systematic and random shifts were (LR), (SI) and (AP) for US, (LR), (SI) and (AP) for CB, and (LR), (SI) and (AP) for SM. The mean 3D shift distance was larger using US compared to CB and SM ( and , respectively). The percentage of US shifts larger than were 34%, 31%, and 38% in the LR, SI, and AP directions, respectively, compared to 18%, 6%, and 16% for CB and 14%, 10%, and 20% for SM. MV-CBCT and SM localization data suggest a different distribution of prostate center-of-mass shifts with smaller variability, compared to US. The online MV-CBCT and SM image-guidance data show that for treatments that do not include daily prostate localization, one can use a CTV-to-PTV margin that is smaller than the one suggested by US data, hence allowing more rectum and bladder sparing and potentially improving the therapeutic ratio.