The aims of this study were to investigate whether intrafraction prostate motion can affect the accuracy of online prostate positioning using implanted fiducial markers and to determine the effect of prostate rotations on the accuracy of the software-predicted set-up correction shifts. Eleven patients were treated with implanted prostate fiducial markers and online set-up corrections. Orthogonal electronic portal images were acquired to determine couch shifts before treatment. Verification images were also acquired during treatment to assess whether intrafraction motion had occurred. A limitation of the online image registration software is that it does not allow for in-plane prostate rotations (evident on lateral portal images) when aligning marker positions. The accuracy of couch shifts was assessed by repeating the registration measurements with separate software that incorporates full in-plane prostate rotations. Additional treatment time required for online positioning was also measured. For the patient group, the overall postalignment systematic prostate errors were less than 1.5 mm (1 standard deviation) in all directions (range 0.2-3.9 mm). The random prostate errors ranged from 0.8 to 3.3 mm (1 standard deviation). One patient exhibited intrafraction prostate motion, resulting in a postalignment prostate set-up error of more than 10 mm for one fraction. In 14 of 35 fractions, the postalignment prostate set-up error was greater than 5 mm in the anterior-posterior direction for this patient. Maximum prostate rotations measured from the lateral images varied from 2 degrees to 20 degrees for the patients. The differences between set-up shifts determined by the online software without in-plane rotations to align markers, and with rotations applied, was less than 1 mm (root mean square), with a maximum difference of 4.1 mm. Intrafraction prostate motion was found to reduce the effectiveness of the online set-up for one of the patients. A larger study is required to determine the magnitude of this problem for the patient population. The inability in the current software to incorporate in-plane prostate rotations is a limitation that should not introduce large errors, provided that the treatment isocentre is positioned near the centre of the prostate.