Abstract This paper presents the numerical results of 6 beam–column joint tests using six types of connections: web cleat, fin plate, top and seat with web angles (TSWA) (8 mm thick angle), flush end plate, extended end plate, and TSWA (12 mm thick angle). Both static and explicit dynamic solvers were employed to overcome the problems of convergence, contact, large deformation and fracture simulations. The finite element models were validated against the test results. It is demonstrated that the finite element analyses give reasonable accuracy compared to the test results. The simulation results indicate that a static solver could predict more accurate simulation results than an explicit dynamic solver. But the problem of numerical non-convergence usually occurs when the static solver is employed to conduct fracture simulations. Complete fracture simulations could only be conducted by the explicit dynamic solver although huge computation resources are required for complicated joint models. In addition, an extensive parametric study was undertaken using these validated models to obtain the rotation capacities of various types of connections under catenary action. Finally, some practical design implications have been drawn up from the parametric study and four new connection acceptance criteria of rotation capacities have been proposed to consider catenary action under a middle column removal scenario. The work shows that current acceptance criteria of rotation capacities for steel joints such as web cleat, fin plate, flush end plate and TSWA connections, are probably too conservative as they only consider pure flexural resistance.