This paper presents an efficient algorithm for developing vehicle structures for crashworthiness, based on the analyses of crash mode, a history of the deformation of the different structural zones during a crash event. It emulates a process called crash mode matching where structural crashworthiness is improved by manually modifying the design until its crash mode matches the one the designers deem as optimal. Given an initial design and a desired crash mode, the algorithm iteratively finds a new design whose crash mode is increasingly closer to the desired one. At each iteration, a new design is chosen as the best among the normally distributed samples near the current design, whose mean and standard deviation are adjusted by a set of fuzzy rules. Each fuzzy rule encapsulates elementary knowledge of manual crash mode matching, as a mapping from the differences between the current and desired crash modes, to the changes in mean and standard deviation for sampling a sizing parameter in a structural zone. A case study on a vehicle frontal crash demonstrated the algorithm outperformed the conventional methods both in design quality and computational time.