Despite documented pitfalls, many geoscientists continue to publish problematic oil–oil correlations based on compound ratios determined from “dead oil” that lacks volatiles. Mixtures of petroleum from more than one source rock pose issues for assigning genetic affinity, proportions of contributing end members, and predicted gas–liquid ratios. Compound ratios mix in nonlinear fashion for end members like black oil versus gas condensate. Correlations and allocations of end-member contributions apply only to the molecular-weight range of analyzed compounds. In this study, biomarker concentrations for four pairs of end-member fluids with gas–liquid ratios from 100 to 50,000 SCF/bbl were mathematically combined to yield four series of binary mixtures in 5% increments. The purpose was to evaluate the reliability of chemometric analyses to assign genetic affinities and to deconvolute the relative contributions of end members to each mixture. Hierarchical cluster analysis of biomarker ratios shows that mixtures having major contributions of gas condensate cluster with those dominated by input from the oil-prone end member, resulting in incorrect genetic classifications. Mixture deconvolution by alternating least squares regression of ratios can seriously underestimate the contribution of the gas condensate or higher gas–liquid ratio oil source, as can alternating least squares regression of concentrations on a whole-liquid basis. A biased view of the contributions from source rocks to mixtures hinders understanding of petroleum systems and distorts expectations of fluid phase and bulk properties. However, alternating least squares regression of compound concentrations on a whole-fluid basis correctly assigns the relative contributions of end members because it includes the gas fraction (C₁–C₅) and the liquids (C₆+).