Using the Algebraic Reconstruction Techniques (ART) a method is introduced for division of the correlation coefficients into subsets allowing the three-dimensional reconstruction to be accomplished on a minicomputer. Results from simulations and experimental phantom data show that ART improves depth resolution compared to back-projection, that under-relaxation produces better images in the case of noisy data, and that the division of the correlation coefficients into subsets has no effect on quality. The images depict the expected resolution degradation in the direction normal to the detector plane due to the limited angular range of projections but yield quantitative results whose relative values are good, even though attenuation is neglected.