3D reconstruction of real world objects is becoming more and more popular among computer graphics and computer vision researchers. One of the more practical approaches of achieving this is called Multi View Geometry, a approach that are using images taken of the real world objects to recreate it.  But it is not always easy to know how to get the desired result out of this approach because there are many variables that affects the reconstructions accuracy, for example the number of images used and the resolution of these images. In this paper a silhouette based 3D reconstruction algorithm is evaluated. Three programs are created each with its own purpose. The first program is used to create as perfect silhouette images as possible in order to get as accurate input data as possible. The second program uses these silhouettes and produces a volumetric reconstruction of the object being reconstructed. The third program creates a polygonal mesh from the volumetric data. The polygonal and volumetric reconstructions are then used when evaluating the visual and volumetric accuracy of the reconstructions. The implemented algorithm is capable of running at real-time or near real-time and produces reconstructions with a high accuracy. It is shown how different silhouette resolutions and the resolution of the volumetric reconstructions influence the accuracy of the created reconstructions. It is also shown that the biggest gain in accuracy related to the number of silhouette images used is gained when increasing from one silhouette and up to ten silhouettes and that by increasing the number of silhouettes more only a low improvement in accuracy is gained.