A GA-based technique for reconstructing plane binary images from a minimal number of projections is presented ;its effectiveness is demonstrated by reconstructing two-dimensional objects from their one-dimensional coded images. The algorithm gets projection data from three different angles. An initial population of strings, each of which contains encoding of an image, is generated randomly. Typical as well as some new genetic operations are performed on every generation of a population to produce a new generation. Results obtained are compared to those obtained by one of the well known iterative algebraic reconstruction techniques, and it was found that the proposed evolutionary method is superior to the algebraic method when the number of projection directions is limited to three.