Engineering optimization is often completely automated after initial problem formulation. Although purely algorithmic approaches are attractive, keeping the engineer out-of-theloop also suffers from key drawbacks. First, problem formulation is a challenging task and a poorly formulated problem often causes extra efforts and extended optimization time. Second, stakeholders may not trust the results of an optimization algorithm when presented without context. This thesis uses information visualization to keep designer inthe-loop during design optimization formulation, modeling, optimization, and result interpretation stages. Parallel coordinates is the central representation used, accompanied by two-dimensional projections for navigation and a scatterplot matrix for overview. Methods are presented to split the design and performance spaces into meaningful regions by clustering and by interaction. A new data-mining technique is also presented to find relationships between black-box constraints to remove redundant and unimportant constraints. A software prototype is developed and successfully applied to an automotive assembly optimization problem.