Designing materials with tailored mechanical properties requires understanding the link between microstructure and fracture toughness. For real materials, which exhibit inherent randomness, this necessitates statistical analyses over ensembles of microstructures. This study employs a recently developed FFT-based phase-field fracture model to analyze the effective fracture toughness of statistically heterogeneous microstructures. Ensembles of two-phase microstructures reconstructed from statistical descriptors are embedded in a novel virtual compact-tension-like specimen design to achieve near-stable mode-I crack propagation. Far-field J-integral evaluation quantifies the impact of heterogeneities on fracture toughness. The framework is verified through systematic convergence studies. Statistical evaluation reveals that microstructural anisotropy strongly influences fracture toughness: material phase elongation perpendicular to the crack path enhances toughness by obstructing crack advancement, whereas alignment parallel to the crack direction promotes crack propagation and lowers toughness. The investigated volume fractions show minimal correlation for the established fracture toughness measure, reflecting the dominance of local topological features over global phase content. Material parameter variations demonstrate that phase contrasts in stiffness and toughness critically influence both the magnitude and variability of fracture toughness. The results provide quantitative structure–property linkages for failure of heterogeneous microstructures and establish a computational framework for data-efficient fracture property prediction. This approach may guide the design of microstructures with enhanced mechanical properties and enables uncertainty-aware inverse design strategies.