The process-structure–property relationship in short-fiber reinforced composites made in Fused Filament Fabrication (FFF) remains inadequately understood, much trial-and-error and extensive testing is required to use these materials for load-bearing applications. As a consequence, the largely empirical design process has hindered the adoption of this technology, notably due to the lack of reliable structural analysis capability. In order to surpass the limitations of mechanical property prediction using simplified artificial microstructures, this work demonstrates the decisive advantage of using geometries obtained directly from imaging of printed specimen instead. The analysis of μCT images is performed via a purpose-built extraction tool called OpenFiberSeg, yielding profound insight into the process-structure relation. The use of real microstructures is shown to considerably improve the mechanical behavior prediction capability via dual-scale FFT-based homogenization, bringing relative error margins below 5%, for full anisotropic description. It also becomes possible to investigate the effect of processing parameters such as nozzle diameter and printing pattern on morphological properties and on mechanical behavior, revealing the magnitude of the spatial variation of local properties. The combination of experimental and simulation enables insight that is not accessible to either alone. Original imaging data and source code are made publicly available.