•Progressive collapse fragility of steel MRFs with semi-rigid bolted connections was studied.•Fragility models were derived and accompanied by tornado diagrams for sensitivity evaluation.•Fibre and component modelling were combined for pushdown-based capacity assessment.•Sensitivity of progressive collapse to design criteria and column-loss scenarios was quantified.•Different fragility levels were established for different analysis scenarios and code provisions. Robustness is a key factor in the safety of structures and structural components or sub-assemblies against progressive or disproportionate collapse, the latter term being classically used to indicate a significant disproportion in size between the initial and final damage configurations. Particularly for a steel moment-resisting frame (MRF), the connections play a pivotal role in the robustness and redundancy of the structural system because an alternative load path could be mobilised when/if the beam-column joints have enough capacity to sustain the abnormal extra-loads developed following the loss or notional removal of a support element as consequence of a triggering event. Although the latter is usually but not exclusively related to terrorist attacks (e.g. faulty construction, uncontrolled gas releases, etc.) and the structural behaviour inherently involves uncertainties in loads and system capacity, progressive collapse has been probabilistically assessed in a few studies, mainly focussed on reinforced concrete structures. In light of this, delving more into probabilistic approaches could be timely, and this paper presents the main results of sensitivity and fragility analysis of steel MRFs featuring top-and-seat with web angle connections, leading to the derivation of both tornado diagrams as well as fragility models. The former ones quantify the progressive collapse sensitivity to design parameters, whilst the latter ones provide the probability of failure as a function of a given intensity measure. Uncertainties in material and geometrical properties of building portfolios were modelled and propagated through Monte Carlo simulation, whereas component-based modelling was integrated with the pushdown methodology for damage analysis of the randomly sampled building realisations. The obtained results show the influence of the studied parameters/random variables on progressive collapse behaviour/fragility and provide insights into the effect of changing the seismic intensity for the design and the column-removal scenario for the assessment of a bolted-angle joint system that has not been probabilistically treated up until now.