Abstract: Background and purpose: Positional uncertainties related to the set-up of the prostate, using internal markers and either 2D–2D or 3D images, were studied. Set-up using direct prostate localization on CBCT scans is compared to set-up using internal markers. Material and methods: 20 patients with prostate cancer were enrolled in the study. After each daily session, a set of 2D–2D and 3D images were acquired. The images isocenter was compared to reference images isocenter. For the set-up error analysis the systematic error, μ, and the set-up uncertainties, Σ and σ, were determined for the translational shift in the three directions, lat, lng and vrt. The set-up errors and uncertainties were calculated in the same way for rotations around the three axes, lat, lng and vrt. Results: Set-up uncertainties were evaluated for four different set-up methods. The systematic error uncertainties were found to be in the range 0.38–1.14mm and for the random error 0.79–1.48mm. For rotations the uncertainties ranges were 0.38–1.59° and 0.91–2.18° for systematic and random uncertainties, respectively. Set-up uncertainties, using internal markers or prostate itself to position the target in the isocenter, were comparable. The correlation between the two methods was better for translational shifts of the isocenter than for rotational shifts. Conclusions: The study shows that the precision of the 2D–2D set-up is equivalent to the precision of the 3D images. It also shows that the soft-tissue based set-up needs 1mm larger set-up margins than the marker based set-up for the prostate patients, when CBCT is used for daily verification of the location of the prostate. [Copyright &y& Elsevier]