Abstract Background A drawback of Onyx, one of the most used embolic agents for endovascular embolization of intracranial arteriovenous malformations (AVM), is the generation of imaging artifacts (IA) in computed tomography (CT). Since these artifacts can represent an obstacle for the detection of periprocedural bleeding, this study investigated the effect of artifact reduction by an iterative metal artifact reduction (iMAR) software in CT in a brain phantom. Methods Two different in vitro models with two-dimensional tube and three-dimensional AVM-like configuration were filled with Onyx 18. The models were inserted into a brain imaging phantom and images with (n = 5) and without (n = 10) an experimental hemorrhage adjacent were acquired. Afterwards, the iMAR algorithm was applied for artifact reduction. The IAs of the original and the post-processed images were graded quantitatively and qualitatively. Moreover, qualitative definition of the experimental hemorrhage was investigated. Results Comparing the IAs of the original and the post-processed CT images, quantitative and qualitative analysis showed a lower degree of IAs in the post-processed images, i.e. quantitative analysis: 2D tube model: 23.92 ± 8.02 Hounsfield units (HU; no iMAR; mean ± standard deviation) vs. 5.93 ± 0.43 HU (with iMAR; p < 0.001); qualitative analysis: 3D AVM model: 4.93 ± 0.18 vs. 3.40 ± 0.48 (p < 0.001). Furthermore, definition of the experimental hemorrhage was better in the post-processed images of both in vitro models (2D tube model: p = 0.004; 3D AVM model: p = 0.002). Conclusion The iMAR algorithm can significantly reduce the IAs evoked by Onyx 18 in CT. Applying iMAR could thus improve the accuracy of postprocedural CT imaging after embolization with Onyx in clinical practice.