In this paper we investigate a systematic procedure for determining frequency-dependent weighting functions for an H2 controller in the frequency domain. Based on experimental results from the system identification of a model structure with an active mass driver, we choose frequency-dependent weighting functions, including filters on an earthquake input model, sensor noise, control output, and control force. These are incorporated into the plant to produce an optimal controller. Combining the weighting functions and filters and comparing the trade-off problem between response reduction and control force, an optimal combination of weighting functions and filters is determined. The performance of the designed H2 controller is evaluated by shaking-table tests of a three-story scaled model with an active mass driver.