Background and purpose: This work evaluates the accuracy of deformable dose accumulation for organs at risk (OAR) in MR-guided prostate SBRT using an anthropomorphic deformable phantom. Materials and methods: Six MR-guided prostate SBRT treatment courses were simulated using volumetric OAR (bladder and rectum) information derived from actual patient data. Deformed OAR contours, geometrical landmarks and GafChromic EBT3 film strips (1.25 × 2.0 cm2) placed at the surface of the OARs were used to validate DIR-based dose accumulation in MRgRT. Two DIR methods were applied: an intensity-based deformation (IB-D) applied to the whole image, and a contour-based deformation (CB-D), resulting in a separate deformation and dose accumulation for each OAR. Dosimetric accuracy was evaluated by quantifying the dose differences, and performing a gamma-index analysis between measured and DIR-derived accumulated dose for both OARs. Geometrical accuracy was assessed by measuring the Dice similarity coefficient (DSC), Hausdorff distance (HDD) and residual distance error (RDE) for all markers at each fraction. Results: CB-D resulted in an average dose deviation from film measurements for rectum and bladder surfaces of 0.6% and 0.3%, respectively. IB-D led to worse results resulting in an overall average dose accumulation inaccuracy of 7.2% and 2.5% for rectum and bladder. CB-D also showed a higher geometrical accuracy than IB-D with significantly higher DSC values and lower RDE and HDD deviations. Conclusion: Empirical validation of dose accumulation in MR-guided SBRT for prostate cancer obtained a good agreement with reference film measurements when using a contour-based DIR approach.