The research presented develops a method for capturing user design preferences interactively and then incorporating these preferences into the multi-objective optimization of long-span trusses. A self-organizing map groups truss designs found in genetic algorithm populations based on information concerning quantifiable truss features. The user interactively selects preferred designs by examining representative truss designs from each group. A hybrid neural network trains on the user selections and is used by the multi-objective genetic algorithm to predict whether a user prefers new trusses in new truss designs. The predicted truss preferences are used in combination with structural design criteria during fitness evaluation, selection, and ranking. The addition of user preferences generates more design alternatives on the Pareto-optimal front and the designs evolved reflect the user preferences in addition to optimizing structural criteria.