Within the civil structural community, nonlinear time history analysis has become a ubiquitous tool to evaluate the structural performance of steel structures when subjected to extreme loadings such as earthquakes, blast, and strong winds. Extreme limit states such as structural instability, local bulking, plastic stress/strain localization in critical regions and structural components can be reliably simulated using current available analysis methods. However, the existing methods cannot reliably model fracture—an extreme limit state which may precipitate structural failure and collapse. Henceforth, on both the structural component and system level, researchers and engineers typically implement a capacity check evaluation approach in which a fracture toughness demand index, calculated based on the predicted continuum stress and strain fields, is checked against a material toughness parameter. Conservatively equating end-of-life (e.g., ultimate failure) to fracture initiation rather than to the onset of unstable crack propagation (e.g., cleavage), such approach disregards the remaining inherit capacity of the steel structure or components—as evident in recent large-scale experimental studies in which the steel components often sustained significant amount of stable ductile crack growth prior to ultimate failure. Clearly, a holistic framework or tool to reliably simulate crack propagation in concert with the global analysis of steel components enables a more realistic assessment of the structural performance of steel structures and designs because it captures the complex interactions between the overall structural response and advancing crack front. Depending on many factors such as the existing numerical tools and associated computational cost, nature of the crack propagation, and the size scale, some numerical frameworks may be more appropriate than others for modeling crack propagation in steel structures. Motivated by this, the scope of this project entails modeling crack propagation in steel structures on three different scales: continuum level, structural component level, and structural frame level. At the continuum level, a novel computational framework is developed and implemented to simulate ductile fracture initiation and propagation. This framework incorporates a local micromechanistic continuum damage model into a cohesive zone model; the continuum damage model predicts fracture initiation, whereas the cohesive zone model simulates the physical process of crack growth and propagation. The framework has been demonstrated to give reliable results (i.e., mesh-convergent agreement between test data and simulations using a single set of model parameters) using test data from CNT and CT specimens. At the structural component level, the framework successfully simulates crack propagation in test specimens that are meant to imitate practical structural design details such as the bolted connections and the reduced-beam-section (RBS) specimen under monotonic loading. Ideally, on the structural frame level, the established framework may be applied to model fracture propagations in key structural components throughout the frame. However, the high computational cost renders such approach impractical. Clearly, a phenomenological frame-element based model is more appropriate. Such model is developed to simulate post-fracture response of welded column splices. The novel model is informed by fracture-mechanics based estimates of splice strength and reproduces phenomena such as gapping and re-seating that occurs in the splices after fracture. Specifically, within the framework of Performance Based Earthquake Engineering, the effects of column splice fracture on the seismic performance of steel moment frames are assessed. It is concluded that due to the rocking phenomenon (e.g., rocking of the top stories above a story with fractured column splices), splice fractures auspiciously affect the dynamic response. Additionally, the phenomenology of splice fracturing throughout the structural system are investigated.