Fracture is an extreme limit state in steel structures, often precipitating structural failure or serious loss of function. Methods to predict fracture in civil structures include traditional approaches developed in other disciplines (mechanical or aerospace engineering) subsequently adopted in structural engineering, as well as approaches to characterize earthquake-induced fractures originating within civil engineering. Developed over nearly six decades, the state of the art is composed of theories and models that address fracture over multiple scales and are targeted toward disparate application scenarios. The paper examines these approaches from a structural engineering standpoint, considering trade-offs in accuracy and expense, while identifying areas for improvement. Traditional approaches (including linear elastic and elastic plastic fracture mechanics) are presented, followed by newer local approaches that are better suited for scenarios where traditional approaches are inapplicable. By simulating micromechanisms such as microvoid growth as well as granular cleavage, local approaches address fracture under large-scale yielding, ultralow-cycle fatigue (which occurs during earthquakes), and low-stress triaxiality, all of which are important in civil structures. The physical basis for these approaches is outlined, with a summary of best practices for calibration and application. However, these local approaches have limitations as well, and often require substantial resources for successful implementation. With this background, optimal fracture assessment strategies are outlined for common structural scenarios, considering accuracy and cost. Limitations of the entire fracture modeling framework are summarized because they pertain to mainstream adoption within structural engineering research and practice. As the profession moves toward accurate performance characterization, it is anticipated that research will accelerate to overcome these limitations.