A methodology for computer simulation of ductile fracture in structural components us- ing the eXtended Finite Element Method (XFEM) is presented. An uncoupled approach is adopted, where plasticity is modeled using the rate-independent J2 flow theory, while damage growth is modeled using a micromechanics-based void growth law. Crack initia- tion at a material point is assumed to occur as a result of a plastic instability, due to loss of ellipticity of the equilibrium equations for a porous material in a state of ‘inhomoge- neous yielding’ by plastic flow localization along bands of voids at the micro-scale. The failure criterion is implemented as a user-defined crack initiation criterion in the commer- cial finite element code, Abaqus. Material separation is modeled using the cohesive zone method, where cohesive elements are dynamically inserted inside elements that satisfy the crack initiation criterion. The method is illustrated by comparing the model predictions with experimental data for specimens made of stainless steel 316LN. Fracture data ob- tained using a range of specimen geometries, including smooth and notched tensile bars, compact tension specimens, and four-point bending of thin-walled pipes with through- wall pre-cracks, are used to validate the predictions of the model. It is shown that the instability-based crack initiation criterion is able to quantitatively predict the strains to crack initiation in the uncracked specimens, and the crack growth rates in the pre-cracked specimens, using a single set of material parameters calibrated from standard tests. In contrast, widely used continuum damage models are unable to predict the response of all specimens using a unique set of parameters.