The presence and form of initial geometric imperfections can significantly impact the structural response and the ultimate capacity of steel structures. Suitably chosen imperfections must be adopted during the design stage, particularly when utilising advanced analysis; applying inappropriate imperfections, either in terms of their shape or amplitude, both of which hold significant influence over the structural response, can cause misleading and possibly unsafe predictions of the ultimate capacity. Some guidance is provided by international standards with regards to choosing imperfection amplitudes, however, aside from suggesting that the imperfection shape should be severe, which is to say that it should cause a large reduction in the ultimate capacity, hitherto it has not been clear how such imperfection shapes should be defined. Therefore, the focus herein is on the development of practical imperfection definition methods that enable severe imperfection shapes to be determined for in-plane stability design of steel moment frames and braced frames via advanced analysis. A series of parametric studies are conducted across a test set of 21 moment frames and 18 braced frames to quantify the effects of initial geometric imperfections on the ultimate capacity. These are used to identify the most severe imperfection shape, thus providing insight and a benchmark against which several proposals for introducing initial geometric imperfections are assessed regarding their ability to reduce the ultimate capacity. In total, over 1.1 million geometrically and materially nonlinear analyses with imperfections (GMNIA) have been conducted using beam finite element (FE) models that incorporate the strain limits from the Continuous Strength Method (CSM). Considerations are also made into the applicability of structural optimisation techniques for identifying severe imperfection shapes. A promising prototype algorithm is devised and implemented which identifies imperfection shapes that perform similarly to the benchmark from running a considerably small number of GMNIA.