Light scattering by single, inhomogeneous mineral dust particles was simulated based on shapes and compositions derived directly from measurements of real dust particles instead of using a mathematical shape model. We demonstrate the use of the stereogrammetric shape retrieval method in the context of single-scattering modelling of mineral dust for four different dust types – all of them inhomogeneous – ranging from compact, equidimensional shapes to very elongated and aggregate shapes. The three-dimensional particle shapes were derived from stereo pairs of scanning-electron microscope images, and inhomogeneous composition was determined by mineralogical interpretation of localized elemental information based on energy-dispersive spectroscopy. Scattering computations were performed for particles of equal-volume diameters, from 0.08 μm up to 2.8 μm at 550 nm wavelength, using the discrete-dipole approximation. Particle-to-particle variation in scattering by mineral dust was found to be quite considerable and was not well reproduced by simplified shapes of homogeneous spheres, spheroids, or Gaussian random spheres. Effective-medium approximation results revealed that particle inhomogeneity should be accounted for even for small amounts of absorbing media (here up to 2% of the volume), especially when considering scattering by inhomogeneous particles at size parameters 3<<i>x</i><8. When integrated over a log-normal size distribution, the linear depolarization ratio and single-scattering albedo were also found to be sensitive to inhomogeneity. The methodology applied is work-intensive and the light-scattering method used quite limited in terms of size parameter coverage. It would therefore be desirable to find a sufficiently accurate but simpler approach with fewer limitations for single-scattering modelling of dust. For validation of such a method, the approach presented here could be used for producing reference data when applied to a suitable set of target particles.