Accurate reproduction of time series with diverse frequency characteristics is a central issue in structural testing. This is true not only for simple experimental tests performed by reaction walls or shaking tables but also for more sophisticated ones, such as hybrid testing. Especially in the latter case, where actual feedback from an ongoing test is used in the calculation of the next excitation value, any possible mismatch may be fatal for both the validity of the test and the safety. The objective of this study is to propose a framework for the adaptive inverse control of shaking tables, which succeeds in this matching to a certain degree. By formulating a critical set of design specifications that correspond to safety, implementation, robustness and ease of use, the conducted research results in a design that is based on a modified version of the filtered-X algorithm with very competitive features. These are the following: (i) default operation in hard real-time and acceleration mode; (ii) very low hardware requirements; (iii) effective cancelation of the shaking table's dynamics; and (iv) robustness against specimen dynamics. For its practical evaluation, the method is applied to shaking table waveform replication tests under the installation of an approximately linear specimen of sufficiently high mass and complex geometry. The results are promising and suggest further research toward this field, especially in conjunction with hybrid testing, as the method retains certain global applicability attributes and it can be easily extended to other transfer systems, apart from shaking tables. A thorough review about actuator dynamics and related control issues (including the delay compensation problem) is given by Plummer [8]. Considering the transfer system in terms of systems theory, an ideal performance is depicted in Figure 1, in which the frequency response function (FRF) of a pure delay is illustrated. If the transfer system could be modeled by such a delay, then the input command (henceforth referred to as the reference signal) would be transmitted to the specimen unaltered (in the steady state), by introducing only an input-to-output delay. Such an assumption may be roughly valid for small-scale transfer systems, yet, this is not the case when complex, large-scale facilities of many mechanical parts are utilized for structural tests. This is particularly true for shaking tables, which require careful fine-tuning in order to take into account all the involved actuators (even when one-dimensional motion is applied), the corresponding kinematics, the resulted eccentricities of the table due to the installed specimen and the dynamics of the specimen itself. In their presentation of the controller for the E-Defense shaking table [9], Tagawa and Kajiwara show two indicative results about the tracking performance in both unloaded and loaded states, from where it follows that the controller approximates the 0-dB gain up to 2 Hz (three translational motions, unloaded case), reaching up to 4 Hz in the loaded case ...