The chain code is an attractive and economical method of coding a binary image, but it is not particularly easy to generate. This article describes a hardware/software mechanism for improving the speed of chain-code generation. At the heart of the procedure is an r.o.m. which is used to detect certain (binary) patterns in the input image. The r.o.m. also controls the acquisition of data by an r.a.m.; the result is a set of data triples of the form (xco-ordinate,yco-ordinate, chain-code element) describing a certain subset of the edge points. A software algorithm is then used to ‘thread’ these elemental items together to generate the complete chain-code string. The procedure can accommodate holes and multiblob images.