In active vibration control, model accuracy of a vibration field is crucial to the stability and performance of closed-loop systems, especially multiple-input-multiple-output feedback control systems. A state-space model is popular for the design of vibration controllers. Its accuracy may be affected by mode truncation, errors in eigenfunctions for a modal model or errors in mass/stiffness coefficients of finite elements for a finite element model. There are few analytical results on controller stability margins with respect to these errors. This paper proposes a controller based on transfer matrices identified from the measurement data, on the ground that the accuracy of transfer matrices is manageable by identification algorithms. The proposed controller is able to introduce active damping to vibration fields. An analytical link is available between the stability margin and identification errors for the proposed controller. These are important features analyzed theoretically and verified numerically and experimentally here.