The results of an ongoing research activity in the development and implementation of reduced-order model-based feedback control of subsonic cavity flows are presented and discussed. Particle image velocimetry data and the proper orthogonal decomposition technique are used to extract the most energetic flow features or POD eigenmodes. The Galerkin projection of the Navier-Stokes equations onto these modes is used to derive a set of ordinary nonlinear differential equations, which govern the time evolution of the modes, for the controller design. Stochastic estimation is used to correlate surface pressure data with flow field data and dynamic surface pressure measurements are used for real-time state estimation of the flow model. Three sets of PIV snapshots of a Mach 0.3 cavity flow were used to derive three reduced-order models for controller design: (1) snapshots from the baseline (no control) flow, (2) snapshots from an open-loop forced flow, and (3) combined snapshots from the cases 1 and 2. Linear-quadratic optimal controllers based on all three models were designed and tested experimentally. Real-time implementation shows a remarkable attenuation of the resonant tone and a redistribution of the energy into various modes with much lower energy levels.