This dissertation composes three papers detailing work intended to examine the mechanisms that control the failure of additively manufactured (AM) metallic materials. All studies aim at developing models to predict the fatigue performance of AM Ti-6Al-4V utilizing computational tools, and ultimately to realize a predictive process-structure-performance modeling framework. First, a thorough literature review was conducted to catalog and analyze the published fatigue performance data of AM Ti-6Al-4V. Comparing to traditionally manufactured Ti-6Al-4V, the key features that control fatigue performance of AM parts are the same as the ones governing traditional cast and wrought, i.e. surface finish, residual stress, build defects and microstructure. Second, possible mechanisms by which hot isostatic pressing (HIP) improves the high cycle fatigue performance of powder bed fused (PBF) Ti-6Al-4V were examined. The results suggest that HIP may act most significantly by decreasing the fraction of the defect population that can initiate fatigue cracks, both by decreasing defect sizes below a threshold and by changing the microstructure around defects, the latter of which is confirmed by an electron backscatter diffraction (EBSD) study. The gained understanding is used to provide initial guidance on the choice of optimum HIP parameter via a continuum mechanics model. Third, to understand the role of geometry on the fatigue performance of as-built laser PBF Ti-6Al-4V, two sets of specimens only differing in gage surface area were fabricated. A significant difference in fatigue performance was observed between these two geometries. Possible origins of the observed difference, such as surface roughness, microstructure, build defects, and mechanical interactions between fatigue cracks and the specimen geometry, were investigated. A weakest link approach with X-ray CT measurements of build defects in the two geometries was able to capture the geometric effect on the fatigue performance, consistent with the experimental data. To attempt to link the defect populations with the thermal histories resulting from the different geometries, a linear heat conduction thermal model was performed. Lastly, a predictive model framework that links the material properties from the witness coupon data with the fatigue performance of structural components was developed to facilitate the qualification and certification of PBF technique in safety critical loading bearing applications. Focusing on a PBF aircraft link component, a simple deterministic model was first developed. A probabilistic fatigue model was then constructed to predict both the median trend and variability of the fatigue life of the link component, taking both the populations of fatigue crack initiation sites and the full stress field from finite element (FE) simulations as inputs. Both models under-predicted the median fatigue life of the link component, which we hypothesize to be due to the overestimation of the multiaxial fatigue indicator. The predicted variability of the fatigue life of the probabilistic model showed good agreement with the experiments. In addition, the effect of crack initiation site density on the fatigue performance of the link component was also examined with the probabilistic model. The results suggested a significant increase in crack initiation site density only has a mild effect on the fatigue performance at the component level, indicating that transferring the fatigue data at the witness coupon level to the structural component level should be conducted more carefully.