AbstractMR image synthesis has previously been developed as a means of retrospectively optimizing contrast of arbitrary materials in MR images. The first step of this process is to form computedN(H),T1, andT2images from source images acquired at a variety of echo delay and repetion times. The second step is to take these computed images, along with operatorselected timing parameters, and mathematically generate a synthesized image. Computation is carried out pixel by pixel according to the equation describing the chosen pulse sequence. This paper presents a study of design considerations for a digital image processor capable of rapidly performing the second step, the actual synthesis. In this work the computations inherent to image synthesis are identified, and the feasibility of performing them in highspeed hardware examined. An analysis of the imprecision due to bit‐limited calculations shows that an error bound of 0.4% is possible with a 16‐bit processor design. A method is described which uses a commercially available image processor by which images can be synthesized according to any of the standard pulse sequences in less than 600 ms. © 1986 Academic Press,