This paper describes an automated approach for optimizing the performance of bipolar ECL circuit. A quadratic equation representing an approximate surface is used to express the circuit delay in terms of the power partition and current densities in the current-switch and the emitter-follower stages. During the iteration of the optimization process, the optimal obtained from the present approximate surface is used as the new nominal point for the next iteration. As the nominal point converges to the optimal, the approximate surface converges to a section of the real optimum surface. This methodology transforms the circuit optimization into a multi-variable optimization problem and is shown to provide an optimum design with circuit analysis accuracy. The design considerations of high-performance ECL circuit are also discussed.