A parallel tempering method was developed for the reconstruction of porous media using information extracted from 2D digitized microscopy images. The method affords significant improvement over simulated annealing, can be easily parallelized, and works generally for isotropic and anisotropic structures. Inclusion of the pore size distribution function in the reconstruction method was decisive for best results. The validity of the results has been checked by comparing cluster size distribution functions, percolation, and transport properties between original and reconstructed images.