<p>The segmentation of 3D surfaces is a crucial process in the analysis and reconstruction of complex objects, especially for cultural heritage artifacts, often fragmented and disseminated. The accuracy of this process also depends on the digitization step, which introduces noise, geometrical or topological distortions, and loss, reflecting the sensor’s limitations and the roughness of the eroded items. Traditional segmentation methods, like DBSCAN Ester (In: Knowledge Discovery and Data Mining, 1996), face challenges when applied to rough, multi-density surfaces due to their limited feature consideration. The proposed Multi-space DBSCAN (MS-DBSCAN) handles multiple surface characteristics in distinct spaces during the propagation process. By exploiting diverse characteristics when identifying neighbors for clustering, MS-DBSCAN improves segmentation quality across varied and challenging datasets, enabling a more precise clustering approach suited to complex surface geometries and decreasing the impact of noise. We validate our method on synthetic and cultural heritage datasets, showing significant improvements in accuracy and robustness over conventional techniques. As the proposed method can handle multiple spaces, it can also be extended, following the needed geometrical properties. Furthermore, we supply an open-source application tool that implements MS-DBSCAN, providing an accessible platform for researchers to perform efficient 3D surface segmentation. Our contributions pave the way for advanced analysis and reconstruction tasks, particularly in the field of cultural heritage preservation.</p>