Anthropogenic releases of petroleum fluids, which contain thousands of compounds, represent a threat to marine environments. However, there was a lack of models able to explain the independent behaviors of the numerous compounds of a spilled petroleum fluid, in particular for a release in deep waters that is subjected to elevated pressures. In this thesis, we present new models to predict the behavior of hundreds of petroleum compounds upon release in the environment, both at the sea surface and in deep waters. We also propose advances to data-analysis techniques. Comprehensive two-dimensional gas chromatography (GC×GC) is used for quantitative and detailed analysis of petroleum composition, enabling the development of models able to predict the independent behavior of numerous petroleum compounds. Nevertheless, uncontrolled run-to-run variations of analyte retention times are encountered in GC×GC chromatograms and this hampers quantitative comparisons. Therefore a new alignment algorithm was developed, enabling improved analysis in subsequent chapters. During the early period after release at sea surface (first hours), several petroleum hydrocarbons fractionate into air and water. However there is a lack of detailed compositional data for this early period, which cannot usually be sampled. We developed a model of evaporation and aqueous dissolution applicable to a whole GC×GC chromatogram and to individual compounds. This model was validated using data recorded previously during a 4.3 m3 oil release experiment conducted on the North Sea. Our model enabled us to investigate the fractionation of hydrocarbons during this early period, and to estimate the expected outcome when several environmental conditions are varied. The thermodynamic properties of petroleum gas and liquid phases released in deep waters are poorly known. To address this need, we present a thermodynamic model of the gas-liquid-water partitioning, densities, and viscosities of petroleum mixtures with varying composition, as a function of pressure, temperature, and water salinity. This enabled us to provide estimates for poorly characterized properties at ambient conditions encountered at emission depth during the 2010 Deepwater Horizon disaster, and to investigate the effect of depth on equilibrium aqueous solubility. The hydrocarbon fractionation reported during the Deepwater Horizon disaster had not been mechanistically explained. To address this need, we developed a model of the combined effects of buoyant plume dynamics and aqueous dissolution kinetics, including relevant deep-water effects. This is the first study to demonstrate mechanistically that aqueous dissolution was a major process, with ~27% of the emitted mass dissolved during ascent. Our model predictions also provide insight in the debate on whether the injection of dispersant at the emission source led to the formation of <300 ¿m droplets that stayed submerged for weeks or months. Biodegradation plays a major role in the natural attenuation of oil spills. However, limited information is available about biodegradation of different saturated hydrocarbon classes, despite that oils are composed mostly of saturates. In collaboration with the Woods Hole Oceanographic Institution, we studied weathered oil samples collected on Gulf of Mexico beaches 12¿19 months after the Deepwater Horizon disaster. We determined the difference in susceptibility to biodegradation for different saturates in the n-C22¿n-C29 range