Abstract: A new algorithm for reconstructing binary section from projections is proposed. Based on the binarity of section to be reconstructed, the algorithm uses pattern recognition instead of estimation for reconstruction. The computer simulation results have shown that the algorithm has following advantages: only a few projections are needed, it can detect fine defects, it still works well in the limited view condition. In this paper the algorithm is described in detail, together with its Neural Network implementation. A method to guarantee the stability of the NN is proposed. A set of simulating results are given.