Abstract: The ductile fracture of metallic materials is usually the result of void nucleation, growth and coalescence. The classical Gurson-Tvergaard (GT) model has proven to be an effective means for simulating the nucleation and growth of voids with homogenous deformations. However, as for the localized deformation due to the void coalescence, the GT model is limited. The original GT damage model is extended to the problem of localized deformations by incorporating two different void coalescence criteria in the paper. One is based on the plastic-limit-load model proposed by Thomason; the other assumes the onset of void coalescence using a critical equivalent plastic strain as a power law of stress triaxiality (defined by the ratio of the hydrostatic stress to the equivalent stress). Hence, void coalescence can be controlled by physical mechanisms, rather than by a critical void volume fraction which cannot be taken as a constant. The extended constitutive models are implemented into an implicit finite element code via a user defined material subroutine (UMAT) in ABAQUS. The void-based meso-damage model was adopted to study the ductile fracture of a series of notched round tensile bars. It is shown that the predictions of the fracture behavior from void nucleation to final material failure have a good agreement with the experiment data, which validates the proposed model. Also, the effects of stress triaxiality on the nucleation and propagation of microcracks are discussed.