This paper proposes a new approach to the problem of producing high quality halftone images. In order to reproduce continuous-tone images on bilevel output devices such as non-impact printers, various halftoning techniques have been developed. Ordered dither and error diffusion (or minimum average error) are the typical methods widely used for the purpose. In these methods, decision of bilevel is made on each pixel bases. In this paper, a new method named “Optimum Pattern Selection” is proposed. In the proposed system, each non-overlapping block of n × n pixels is taken up from the original continuous-tone image. On the other hand, all the possible binary patterns of the same size, i.e. n × n pixels, are generated mathematically. Each of the binary patterns is then transformed to a reproduced continuous-tone image block by a lowpass filter, which is a simplified model of HVS (Human Visual System). The reproduced images are then compared to the original image block. The optimum binary pattern, which brings about the reproduced image most similar to the original image block, is selected. For the purpose of improving the halftone image quality further, the optional filter switching algorithm is introduced.