The purpose of the study was to evaluate accuracy of patients' set up verified by kV CT-on-rails system and compare automatic and manual image registration of planning and verification kVCT-s. Between January 2001 and March 2011, at ten patients with prostate cancer the clinical target volumes (CTVs) for prostate (CTV-PROS), and prostate plus caudal 1 cm of seminal vesicles (CTV-PVS) with or without pelvic lymph node region were contoured on the treatment planning CT, according to risk category of the patient. Planning target volumes (PTVs) were created with 1 cm margin extended around the CTVs in each direction. The isocentre was marked on the skin with three radiopaque markers. After the set up, treatment couch with the patient was turned by 180 degree and images were acquired of the region of the isocentre with a kV helical CT-on-rails system (treatment CT). An image registration software was used to co-register planning and treatment CT images. Automatic CT image co-registration was followed by manual co-registration taking into account the CTV-PROS contour and soft tissue informations. Deviations of the isocentres in lateral (LAT), longitudinal (LONG) and vertical (VERT) directions were recorded after each image co-registration. Corresponding data were compared using the t-probe. Systematic (S) and random (s) errors of the set up were calculated. Adequate PTV to CTV margins were calculated by van Herk's formula (2.5xS + 0,7xs). Overall 252 deviations were analysed on fourty-two CT series of 10 patients. The mean errors of the set up with automatic and manual image co-registrations were 0.19 cm and 0.07 cm (p=0.001) in LAT, 0.05 cm and 0.03 cm (p=0.07) in LONG and 0.16 cm and 0.22 cm (p=0.16) in VERT directions, respectively. The systematic errors of the set up for automatic and manual image registrations were 0.22 cm and 0.26 cm in LAT, 0.17 cm and 0.18 cm in LONG, 0.25 cm and 0.26 cm in VERT directions, respectively. The random errors of the set up for automatic and manual image registrations were 0.31 cm and 0.26 cm in LAT, 0.27 cm and 0.27 cm in LONG and 0.24 cm and 0.33 cm in VERT directions, respectively. In case of manual image co-registration, the required PTV to CTV margins to cover at least 95% of the CTVs with at least 95% percent of the prescribed dose were calculated to 0.93 cm in LAT, 0.65 cm in LONG, and 0.89 cm in VERT directions. Patients set up can be verified with manual image co-registration based on soft tissues around the prostate using a kV CT-on-rails system installed in the treatment room. The difference between automatic and manual image co-registration was significant in LAT direction. A PTV to CTV margin <1 cm seems to be appropriate to cover the CTVs in image-guided prostate radiotherapy. These findings support our recent clinical protocol.