In this paper, a simulated annealing approach is used to solve the left ventricle, tridimensional reconstruction problem based on two angiographic views. The algorithm works under the assumption of having parallel projections and a homogeneous mixture of blood and contrast agent filling the ventricular cavity. Those assumptions allow to decompose the original 3D object into a stack of bidimensional slices, in order to simplify the reconstruction processo Each slice is modeled as a bidimensional Markov-Gibbs random field and the reconstruction process is started with an initial approximation, which is appropriately deformed by using a simulated annealing optimization procedure in order to minimize an energy function that includes projection compatibility and spatial regularity constraints. The results obtained by reconstructing a 3D binary database of the ventricle from two synthesized projections, as well as the reconstruction obtained from two orthogonal angiographic images show a low reconstruction error and the shapes of the reconstructed slices are highly correlated with the original database.