Numerous studies indicate the need to account for particle non-sphericity in modeling the optical properties of dustlike aerosols. The methods for simulating the scattering of light by various non-spherical shapes have rapidly evolved over the last two decades. However, the majority of aerosol remote-sensing retrievals still rely on the Mie theory because retrievals accounting for particle non-sphericity are not as well defined methodologically and are demanding computationally. We propose a method for the retrieval of the optical properties of non-spherical aerosol based on the model of a shape mixture of randomly oriented polydisperse spheroids. This method is applied to angular and spectral radiation measurements from the Aerosol Robotic Network (AERONET) in locations influenced by desert dust. Comparisons with Mie-based retrievals show a significant improvement in dust-particle phase functions, size distributions, and refractive indices.