Optical mapping allows the visualization of cardiac ac-tion potentials on cardiac tissue surfaces by fluorescence using voltage-sensitive dyes. So far, the surface measure-ments are directly related to surface action potentials. We aim at exploiting these measurements in order to recon-struct three-dimensional action potentials. In this prelimi-nary work, we tried to localize fixed electrical sources in a three-dimensional slab of tissue. Therefore, we minimized the difference between the measure and a model, in a least squares sense. The model is based on the diffusion equa-tion for the propagation of light, and simplified spherical or ellipsoidal electrical wavefronts. The inverse problem is solved by combining the gradient method and the BFGS algorithm. The method was challenged on data in silico, and then on experimental data. The results show improve-ments over previous localization methods on data in silico, although its use on experimental data still requires some work.