Extracellular vesicles (EVs) are generating a growing interest because of the key roles they play in various biological processes and because of their potential use as biomarkers in clinical diagnostics and as efficient carriers in drug-delivery and gene-therapy applications. Their full exploitation, however, depends critically on the possibility to classify them into different subpopulations, a task that in turn relies on efficient means to identify their unique biomolecular and physical signatures. Because of the large heterogeneity of EV samples, such information remains rather elusive, and there is accordingly a need for new and complementary characterization schemes that can help expand the library of distinct EV features. In this work, we used surface-sensitive waveguide scattering microscopy with single EV resolution to characterize two subsets of similarly sized EVs that were preseparated based on their difference in buoyant density. Unexpectedly, the scattering intensity distribution revealed that the scattering intensity of the high-density (HD) population was on an average a factor of three lower than that of the low-density (LD) population. By further labeling the EV samples with a self-inserting lipid-membrane dye, the scattering and fluorescence intensities from EVs could be simultaneously measured and correlated at the single-particle level. The labeled HD sample exhibited not only lower fluorescence and scattering intensities but also lower effective refractive index (n approximate to 1.35) compared with the LD EVs (n approximate to 1.38), indicating that both the lipid and protein contents were indeed lower in the HD EVs. Although separation in density gradients of similarly sized EVs is usually linked to differences in biomolecular content, we suggest based on these observations that the separation rather reflects the ability of the solute of the gradient to penetrate the lipid membrane enclosing the EVs, that is, the two gradient bands are more likely because of the differences in membrane permeability than to differences in biomolecular content of the EVs.