To fully advance our understanding of hydrocarbon conversion chemistry requires powerful analytical methods to qualitatively and quantitatively characterize complex petroleum fractions at the molecular level. In the absence of such tools, an alternative solution is to model the molecular composition of hydrocarbon mixtures with limited analytical data. The objective of this study is to integrate modeling techniques with conventional and advanced petroleum characterization methods to derive the composition of middle distillate fractions at the molecular level. In the present approach, analytical petroleum characterization data are used as input to computationally generate a mixture of representative molecules that mimics the properties of the real sample. The representing molecules are constructed according to coherent chemical/thermodynamic criteria by Monte Carlo sampling of a set of statistical functions assigned to each possible molecular feature. The assembled mixture is built on a large set of chemical species and is further optimized with the principle of Maximum Entropy. The approach is applied to simulating two middle distillates differing significantly in hydrocarbon type composition and origin. The samples are experimentally characterized by standard and advanced analytical methods: density, simulated distillation, elemental analysis, hydrocarbon types/distributions and sulfur compound speciation by two-dimensional gas chromatography with flame ionization detector (GC x GCFID) and sulfur chemiluminescence detector (GC x GCSCD), and C-13 nuclear magnetic resonance (NMR), to obtain sufficient information for parameter fitting and model validation. Simulation results showed that the model is capable of generating representative mixtures that reasonably match the actual physical samples in analytical properties and carbon number distributions.