Chemometric tools (partial least squares, PLS) with variable selection (iPLS, siPLS, UVE, and GA), associated with near-infrared (NIR) and mid-infrared (MIR) spectroscopy, were applied for the determination of sulfur content (weight percent) in petroleum derivatives. Evaluation of the models was conducted by determination and analysis of the following requirements: coefficient of determination (R-2), the curve obtained by plotting predicted versus measured values, and also cross-validation and prediction errors. The developed models presented in this work had satisfactory results, demonstrating that both NIR and MIR techniques, combined with chemometric tools, can be used to determine the sulfur content in petroleum fractions, with limits of detection and quantification (LOD and LOQ) from 0.0234 and 0.0781, respectively. All variable selection techniques improved the predictive ability of the mode in comparison with the full-spectrum model, except the UVE method, which showed better performance only when associated with MIR data.