The demand for improved gasoline product quality has helped make molecular-level models become more preferred for the modern refinery. Building the molecular compositional model is an essential first step for this quantitative molecular management of gasoline streams. Gas chromatography equipped with flame ion detection (GC-FID) is commonly used in the gasoline detailed hydrocarbon analysis (DHA). The combination of GC-FID analysis and molecular-level modeling is thus very attractive. In the present study, we developed a gasoline compositional model based solely on GC-FID as input. To suppress the negative influence of peak coelution, we developed a statistics-based peak tuning algorithm to obtain individual compound resolution at higher carbon number range. Using the tuned result as input, the molecular-level gasoline compositional model was built by estimating the quantitative percentages of the species in a predefined molecular library (573 molecules). The molecular-level compositional model has good extensibility and can link to the molecule-based physical properties prediction model. The model has been verified via applications on various gasoline samples. The prediction of research octane number for large-scale gasoline samples was also revealed.