Particle roundness has been proved to have a considerable influence on the macroscopic and microscopic mechanical properties of granular particle assemblies. Characterizing roundness by defining a quantitative shape descriptor is of great importance. Due to the geometrical complexity of realistic particles, it remains a considerable challenge to establish a 3D quantitative roundness descriptor. In this context, a robust numerical approach is developed for quantitative and precise computation of a roundness descriptor for granular particles. Corner identification and sphere-fitting algorithms are developed to compute the 3D roundness descriptor Rd(3D). Finally, the proposed method is applied to different granular particles, e.g., superball particles, bank gravel and crushed rock. The results indicate that the proposed Rd(3D) descriptor can provide reasonable and robust evaluations of particle roundness. (C) 2018 Elsevier B.V. All rights reserved.