The previous characterization of petroleum by real-time analysis is a need in many refining plants and is considered a complex challenge because of the feedstock characteristics, which can vary over a widely extended range. Typical examples consist of petroleum analysis for the determination of quality parameters, such as true boiling point (TBP) distillation curve and American Petroleum Institute (API) gravity, which are excessive time-consuming analysis when traditional standard methods are used. A multivariate calibration using spectroscopy data combined with chemometric methods can be used to overcome this drawback, but in calibration problems involving analysis of complex mixtures, such as petroleum and its derivatives, there is a difficult to reproduce composition variability of real samples by means of optimized experimental designs and a chemometric method with a good generalization performance is required. The algorithm support vector machines (SVM) is able to adequately treat nonlinear relationships and with high generalization performance, providing in many cases better results than traditional partial least-squares (PLS) regression models. This work demonstrates the development of an innovative method based on the fast energy-dispersive X-ray fluorescence and scattering spectroscopy and chemometric methods, such as SVM and PLS, aiming to determine TBP curve and API gravity of crude petroleum samples. Many advantages can be attributed to the developed method when compared to reference method determinations, such as low cost and mainly higher speed, which are of great interest for process optimization activities.