A genetic algorithm based approach is employed in the inverse problem of reconstructing the interior of a diffusing medium. Diffusion of optical energy in a scattering medium is simulated by a relaxation scheme. The genetic algorithm uses an error measure to successively modify an initial set of solutions yielding new generations of improved solutions. The error measure, which determines the relative merit of a particular solution, is determined by comparing the data obtained by simulating diffusion through the solution with those for the unknown medium. Unlike conventional iterative schemes, successive generations of solutions are generated through a directed parallel search in the solution space without any knowledge of the derivative of the error surface. The parallel search mechanism alleviates the problem of getting trapped in local minima. Results of experiments performed on two-dimensional planar media are presented along with suggestions for hybrid approaches that incorporate other reconstruction schemes.