We present a method for the semi-automatic, model-driven segmentation of the Left Ventricle (LV) in cardiac SPECT data. Accurate segmentation of the LV allows computation of quantities that describe the degree of a perfusion defect. A 3D surface is initialised in the dataset and deforms to fit the shape of the LV. The model is chosen from a small number of prototypes. The behaviour of the dynamic model is controlled by an Active Surface which is simulated by using the Finite Element Method. Segmentation results were obtained for nine datasets for which a manual segmentation by an expert existed. Results show a good agreement with the manual segmentation with an average of the mean contour distance of 0.24 voxel. Heart diseases like myocardial infarcts are common causes of death in the western world. To find and evaluate heart defects, nuclear medicine imaging techniques such as SPECT are used. SPECT as a three-dimensional functional imaging technique may be used to show the amount of perfusion in the parts of the Left Ventricle (LV). The volume of the non-perfused tissue of the LV indicates severity of the malfunction. To perform a quantitative perfusion analysis, it is necessary to incorporate anatomical knowledge into the functional data in terms of a segmentation of the LV. This is challenging because of the poor image quality in SPECT with its low spatial resolution and its low signal-to-noise ratio. Furthermore, clinically relevant datasets have perfusion defects at different positions of the LV. These defects are visible as low-intensity gaps in the images, but have to be included in the segmentation. Hence, we use a deformable model for segmentation that represents the shape of the LV and is registered with the SPECT data. The deformable model contains the necessary anatomical knowledge to evaluate the relation between perfusion and anatomy. Starting from an initial shape that is positioned near to the true solution, the model is adapted to the data. We choose a surface model that represents the shape of the wall of the LV.