Abstract: In reduced-scale modal hybrid simulation, the constraints of test site and of modal construction pose challenges in achieving a fully analogous scaled model, resulting in unavoidable non-exactly similarity error. Consequently, A dynamic Kriging model-based error prediction and control approach are proposed to optimize the design of hybrid simulation scale model and to enhance the precision of hybrid simulation. Taking the 1/2 scaled steel frame model as an example, the maximum absolute error between the predicted and true values does not exceed 5% for different boundary conditions. It is shown that all the dynamic Kriging models have high prediction accuracy. The non-exactly similarity error is controlled under 45%, which is shown to be well achieved after factor values being taken in the optimized design space. Error control was achieved after using the optimal combination of parameter values, and the non-exactly similarity error was reduced to less than 35%. It establishes the groundwork for designing the reduced-scale model hybrid simulation.