In recent years, there is an increase in customer requirement on the comfort of vehicles. As a result, reducing the odor inside the vehicles has become an important and elusive task. Extensive experimental results show that the odor inside vehicles mainly comes from VOC (volatile organic compounds) emitted by the interior ornaments and parts. Given there are many VOC components affecting the odor, determining which VOC components are essential to the odor becomes a main difficulty in optimizing the odor in vehicles. In this paper, we proposed a new approach to optimize the odor of VOC in vehicles based on data-driven modeling and goal programming. To this end, we first collected mass spectrograms of vehicle parts and their odor ratings, where the mass spectrograms are obtained by mass spectrometer and ratings are scored by olfactory engineers. Then we used these data to build a data-driven model based on Weber-Fechner Law. The data-driven model is solved using lasso regression. Based on the data-driven model, we found out the contributions of the VOC components to the odor rating, which enables us to focus on certain specific VOC components that contribute much to the odor ratings. By strategically reducing those specific VOC components using goal programming, we finally obtained an optimized design with a better odor rating. To be specific, when performing the optimization, instead of minimizing the VOC odor rating, we set an ideal odor rating as the goal and formulated the optimization as a goal programming problem. To validate our approach, we collected 179 VOC mass spectrograms to train and test our data-driven model. The average accuracy of predicting odor ratings from mass spectrograms can reach 85% ∼ 90%. This data-driven model implies the contributions of VOC components on different mass weights to the odor rating, and the selected high-contribution mass weights can give reasonable optimization scheme to reduce the VOC gas odor.