Computational fluid dynamics (CFD) is applied to develop a novel submicron air classifier. Based on various sizes and positions of the inner structure of the elbow-jet classifier, its two-dimensional airflow field has been simulated by the Fluent software. For this typical cross-flow field, the standard k-epsilon turbulence model is applied. The Coanda effect plays a paramount role in separating ultrafine particles in the high-speed flow field of the elbow-jet classifier. The factors that influence the performance of the Coanda block, i.e., a quarter-cylinder centered in the classifier, are analyzed and discussed. The trajectories of moving particles with different diameters in the channels and chamber of the classifier have been calculated through the velocity field obtained from the CFD simulation. The cut sizes of three products from three outlets of the classifier are obtained based on the calculations of the trajectories of particles and are compared with corresponding experimental results. The ground and classified experiments have been studied simultaneously where the product outlet of a vortex jet mill is used as the feed in an elbow-jet classifier. The combination of the vortex jet mill with the external classifier provides an alternative to grinding equipment for multiple-size products of fine/medium/coarse powder. A centrifugal channel is added between the vortex jet mill and elbow-jet classifier to improve the performance of the air classifier. Both numerical and experimental results show that the pre-distributed feed particles at the exit of the centrifugal channel have a considerable effect on the separation of fine powder and a lesser effect on the separation of coarse powder.