In this study, crashworthiness of a bus bumper system with a special honeycomb beam is optimized under impact loading using a multi-fidelity optimization approach. The crash performance of the bumper system is evaluated using two metrics: crush force efficiency (CFE) and specific energy absorption (SEA). An optimization with aggregated objectives is performed to seek for an optimum bumper design. Optimum values of the crashbox length, honeycomb wall angle and honeycomb wall thickness are obtained to maximize composite objective function that provides a compromise between these two metrics. Commercial finite element software LS-DYNA is used to compute CFE and SEA values. Multi-fidelity modeling is used to combine data of low-fidelity model at all training points with high-fidelity data at some randomly selected training points to obtain accurate response predictions in less computational time. It is found that multi-fidelity optimization can reduce the computational cost by 33% with only 2% smaller composite objective function value compared to the high-fidelity optimization alternative.