Recently, modeling methods using dynamic measurement data have attracted much attention in the automotive industry. A mathematical model of the gasoline engine is constructed in order to express and predict its transient response during running. Since in the modeling process, we need to handle an extremely large data set, the computational cost increases accordingly. In this paper, we apply the approximated kernel method to combustion modeling in the gasoline engine. An approximation of the kernel reduces the computational cost and enables us to handle a large data set. In addition, a strategy for setting tunable parameters is proposed by analyzing their effect on the model accuracy. The effectiveness of the approach is illustrated by constructing a combustion model using experimental data.