The majority of previous quasi-static tests on disproportionate collapse simulated the column removal through applying concentrated load/displacement on the top of the removed column until failure. However, uniformly distributed service load always exists on the frames. Therefore, to reflect the actual load condition more accurately, uniformly distributed load should be applied along the beams first. Then, the temporary support is gradually removed to simulate the process of column removal. By this way, the beams may undergo only a small deflection as the dynamic effect is neglected. Thus, a subsequent concentrated loading process is employed to evaluate the behavior of the beams at the ultimate stage, which may be reached if the dynamic effect is considered. Such a loading process is named sequential loading regime. To evaluate the effects of loading regimes on the behavior of reinforced concrete (RC) frames under a middle column removal scenario, two series of half-scale RC beam-column sub-assemblages were tested in this study. It is found that the conventional concentrated loading regime could accurately estimate the yield strength and the compressive arch action capacity of the RC beam-column sub-assemblages, but it may over-estimate the catenary action (CA) capacity and the deformation capacity. Moreover, although the concentrated loading regime is convenient and able to demonstrate the load transfer mechanisms of the sub-assemblages against disproportionate collapse, it may mistakenly identify the locations of critical sections. Furthermore, based on the failure modes and local strain gauge results, analytical models were proposed for predicting the CA capacity of the tested specimens under two loading regimes. Results suggest that the analytical models could predict the CA capacity well.