The fracture toughness of a material depends upon the material's composition and microstructure, as well as other material properties operating at the continuum level. The interrelationships between these variables are complex, and thus difficult to interpret, especially in multi-component, multi-phase ductile engineering alloys such as a/b-processed Ti-6Al-4V (nominal composition, wt pct). Neural networks have been used to elucidate how variables such as composition and microstructure influence the fracture toughness directly (i.e., via a crack initiation or propagation mechanism)-and independent of the influence of the same variables influence on the yield strength and plasticity of the material. The variables included in the models and analysis include (i) alloy composition, specifically, Al, V, O, and Fe; (ii) materials microstructure, including phase fractions and average sizes of key microstructural features; (iii) the yield strength and reduction in area obtained from uniaxial tensile tests; and (iv) an assessment of the degree to which plane strain conditions were satisfied by including a factor related to the plane strain thickness. Once trained, virtual experiments have been conducted which permit the determination of each variable's functional dependency on the resulting fracture toughness. Given that the database includes both K 1C and K Q values, as well as the in-plane component of the stress state of the crack tip, it is possible to quantitatively assess the effect of sample thickness on K Q and the degree to which the K Q and K 1C values may vary. These interpretations drawn by comparing multiple neural networks have a significant impact on the general understanding of how the microstructure influences the fracture toughness in ductile materials, as well as an ability to predict the fracture toughness of a/b-processed Ti-6Al-4V.