You're probably thinking of strain hardening, in which some plastic deformation is allowed for prior to collapse. We are essentially relying on the fact that steel is a ductile material, and acknowledge that steel members will not instantly fail as soon as its elastic limit is reached. Note that for some very thin plates in compression, they may fail below the elastic limit due to local buckling, however it is common (at least in the UK) to take this into account using a reduced section modulus rather than factor the actual yield stress.
For S355 steel for example, the elastic yield stress is 355N/mm2 depending on plate thickness. Beyond this point it has reached its elastic limit, and in a linear analysis would be inadequate. If we take into account strain hardening, we can use a slightly higher yield stress. Details of a suitable non-linear stress/strain curve are given in BS EN 1993-1-5, which if I recall allows an ultimate strain of 5% with a gradient of E/10 after elastic yield has occured.
The improved yield stress would thus be = sigma_el + (e_max - e_elastic)*E/10. e_elastic = E/sigma_el