Gas-solid flows in nature and industrial applications are characterized by multiscale and nonlinear interactions that manifest as rich flow physics and pose unique modeling challenges. In this article, we review particle-resolved direct numerical simulation (PR-DNS) of the microscale governing equations for understanding gas-solid flow physics and obtaining quantitative information for model development. A clear connection between a microscale realization and meso/macroscale representation is necessary for PR-DNS to be used effectively for model development at the meso- and macroscale. Furthermore, the design of PR-DNS must address the computational challenges of parameterizing models in a high-dimensional parameter space and obtaining accurate statistics of flow properties from a finite number of realizations at acceptable grid resolution. This review also summarizes selected recent insights into the physics of momentum, kinetic energy, and heat transfer in gas-solid flows obtained from PR-DNS. Promising future applications of PR-DNS include the study of the effect of number fluctuations on hydrodynamics, instabilities in gas-solid flow, and wall-bounded flows.