A case study for improving the quality of a wave-soldering process that produced printed circuit boards (PCB's) is presented. A mixed-level fractional factorial design was implemented in a high-volume production system during normal operational hours. The observed ordered-categorical data from the bottom-side soldered leads were weighted to formulate the average, spatial uniformity, and dispersion process performance measurements. For lead classes like the integrated circuit and printed grid array, a polynomial model was established using the least squares method with weights provided by a dispersion function. The main-effect and interaction model terms were selected by forward and all-subsets regressions. Production quotas, topside defects, presoldering board temperature, and variance models were used to set the constraints for simultaneously optimizing predictions from the average and the uniformity models of all leads. A nonlinear optimization routine was used to determine the best and most robust settings for the continuous and discrete process variables. Results from a confirmatory experiment showed an improvement of mean soldering quality by 33% and of uniformity by 39%.