This paper examines the seismic performance of steel buildings with alternative framing systems subjected to bi-directional ground-motion. Peak drifts of one-way (perimeter framing) and two-way (space framing) systems are assessed by means of scalar and vector-valued probabilistic methods. Extensive non-linear response history analyses over idealized 3D structures representing 6- and 9-storey buildings are performed under pairs of linearly scaled ground-motions. Both far-field and near-field non-pulselike acceleration series are considered. The spectral acceleration of the geometric mean of the two horizontal components (Sa,GM) is taken as the primary intensity measure (IM) while four other ground-motion parameters are employed to construct IM-vectors including: the spectral acceleration ratio (RT3,T1), the spectral shape parameter (Np), and two frequency content parameters (Tm and To). This paper shows that incorporating the vector 〈Sa,GM,Np〉 into the assessment of bi-directionally loaded 3D buildings yields up to 40% lower conditional standard deviations than a purely scalar formulation at large drift levels while the vector 〈Sa,GM,RT3,T1〉 is more efficient at smaller drifts. The effects of alternative framing systems on structural fragilities are found to differ depending on the number of storeys. For 6-storey structures, consistently higher capacities are observed in two-way layouts with respect to one-way systems but they are associated with increasing variabilities at larger demand levels. Conversely, the 9-storey two-way building experiences 5% lower mean capacities than its one-way counterpart. Finally, drift hazard curves are calculated by combining the building fragilities with idealized ground-motion hazard estimates. The results indicate that one-way buildings experience consistently lower drift exceedance rates regardless of the ground-motion type, especially for drift levels larger than 2% although the differences are larger for the 9-storey frames in comparison with their 6-storey counterparts. This study represents a first attempt to implement vector-valued analysis in the context of bi-directionally loaded structures and its results constitute an important step towards discerning the most favourable framing system at different seismic performance levels.