When a structure undergoes seismic excitation, the intensities and spatial distributions of the reactive weights on the structure may not be the same as those assumed in original design. Such a difference is inevitable due to many facts with the random nature (e.g., redistribution of live load), resulting in accidental eccentricity and consequently torsional response in the system. The added torsion can cause excessive deformation and premature failure of the lateral force resisting system. Its detrimental effect is typically accounted for in most building design codes with an arbitrarily specified accidental eccentricity value. While it tends to amplify drift response of buildings under earthquake excitations, it is unclear whether the code specified accidental eccentricity is quantitatively adequate or not in seismic fragility assessment of steel moment frames (including low-rise, mid-rise and high-rise frames) with random reactive weight distributions. This thesis applies surveyed dead and live load intensities and distributions to three representative steel moment resisting frame structures that have been widely investigated in a series of projects under the collaboration of the Structural Engineers Association of California (SEAOC), the Applied Technology Council (ATC), and Consortium of Universities for Research in Earthquake Engineering (CUREE), known as SAC. Based on an extensive parametric study and incremental nonlinear dynamic analyses, it is found that variable load intensity and eccentricity had negligible impacts on the inter-story drifts of the low-and high-rise steel moment frames. However, they affect to a higher degree the performance of the mid-rise steel moment frames. Moreover, it is found that under the maximum considered earthquake (MCE) event, the actual drifts in steel moment frames with random reactive weight distributions can be conservatively captured through consideration of the code specified accidental eccentricities.