Abstract: A scaled hybrid simulation method based on the classification similarity law is proposed to deal with the problem that the scale model and the prototype structure are difficult to meet the complete similarity in scaled hybrid simulation, and the incomplete similarity error is accumulated and amplified in the iterative coupling process. Based on the basic dimension, the parameters are divided into key parameters, basic parameters and derivative parameters. The scaled relationship is constructed through a three-step method to ensure the similarity of key responses. Taking a five-story two-span steel frame as the research object, two constraint cases and six incomplete-similarity conditions are set up to carry out virtual hybrid simulation. Results indicate that, under displacement–restoring force constraints, the displacement error in each working condition is maintained below 1%, with an average error of 0.4%, which is 18.3% lower than that obtained using the traditional dimensional analysis method. The error distribution is also narrower. When stress constraints are added, the stress error stays below 1% in all conditions. Across the six working conditions, the difference between the mean restoring-force errors of the two cases does not exceed 0.2%. These results demonstrate that the method can effectively suppress the error propagation and accumulation in scaled hybrid simulation, and has good scalability and robustness.