Online or real-time system identification techniques are advancing in civil and structural engineering, particularly for the development of emergency response systems and autonomous damage assessment of civil structures. From the perspective of civil and structural engineering, online system identification can be employed to monitor structures’ modal parameters, issue timely warnings regarding potential structural damage, and facilitate post-earthquake reconnaissance and rehabilitation. To enable online system identification, several techniques have been adapted into recursive formulations, including the recursive AutoRegressive eXogenous (ARX) model, recursive instrumental variable methods, and the recursive prediction error method. Recently, subspace identification (SI) has been modified to create recursive SI (RSI), which is capable of detecting time-varying dynamic characteristics of systems. This study introduces various recursive formulations for SI and provides a detailed discussion of the user-defined parameters utilized in these techniques, examining the impact of these parameters on the identification results. To verify the proposed user-defined parameters in RSI, two distinct datasets are employed: the first consists of laboratory tests conducted on a four-story steel frame with variations in column stiffness on the first floor, while the second comprises seismic responses obtained from a real building monitored by the Taiwan Strong Motion Instrumentation Program (TSMIP). The first dataset serves to illustrate the process of selecting an appropriate system order for online system identification techniques. Additionally, the second dataset is utilized to explore the application of forgetting factors to diminish the influence of past data for damage detection. In conclusion, this study provides a summary and general recommendations for selecting user-defined parameters in RSI, culminating in a comprehensive conclusion.