Hydraulic conveying of graded particles is much more complex than that of uniform particles but is not fully understood. A fully coupled computational fluid dynamics–discrete element method model is established for the hydraulic conveying of graded particles, which integrally considers the particle–fluid and particle–particle interactions and turbulence modulation from particles. The proposed model accounts for the stochastic motion of particles by the discrete random walk method and applies the diffusion averaging algorithm to obtain particle concentration in arbitrary cells for smooth and cell-independent data fields on unfavorable cells (fluid cell size ≤ particle size). The particle–fluid drag force is applied to slurry flows through densely packed particle beds due to the consideration of porosity modification. The proposed model well performs in simulating the hydraulic conveying of dense graded particles. Dynamics in slurry mixtures of bi-disperse particles are investigated regarding different particle size compositions. The results show obvious stratification between coarse and fine particles, e.g., fine particles settling at the pipe bottom elevate the coarse particles and form a “lubrication layer” with high velocity. The torque caused by particle–particle/wall contact is greater than the torque caused by the fluid. The pipe cross section is divided into four regions according to the particle angular velocity. The effect of particle concentration on liquid motion is small because the difference in local particle concentration is relatively small, but the maximum pressure drop corresponds to a critical particle size composition.