We investigate the influence of annealing on the morphology of intimately mixed blends of the conjugated polymers poly(9,9'-dioctylfluorene-co-bis-N,N'-(4-butylphenyl)-bis-N,N'-phenyl-1,4-phenylene-diamine) (PFB) and poly(9,9'-dioctylfluorene-co-benzothiadiazole) (F8BT) with scanning transmission x-ray microscopy (STXM). Through the use of a zone plate with theoretical Rayleigh resolution of 30 nm, we are able to resolve sub-100 nm bulk structure in these films. Surprisingly, for unannealed films spin-coated from chloroform we observe features with an average diameter of 85 nm. The high degree of photoluminescence quenching in these as-spun films (>95%) implies that there is significant intermixing within the 85 nm structures, indicating that a hierarchy of phase separation exists even on the length scale of less than 100 nm. With annealing up to 160 °C, close to the T(g) of the components, there is little change in the feature sizes observed by STXM, although an increase in variation of the composition is observed. With annealing above 160 °C the imaged features begin to evolve in size, increasing to 225 nm in extent, alongside large changes in composition with annealing to 200 °C. Comparing the evolution of morphology imaged by STXM with the change in photoluminescence quenching with annealing, we propose that phase separation first evolves via the evolution of relatively pure phases on the length scale of a few to tens of nanometres within the larger 85 nm structures. Once the length scale of compositional fluctuations exceeds 85 nm (for anneal temperatures above 160 °C) the hierarchy of phase separation is lost and the subsequent morphological evolution is readily imaged by STXM. Applying the results of an exciton diffusion and quenching model, we find good agreement between the size of the domains measured by STXM (above 180 °C) and the results of the model for an exciton diffusion length of 15 nm. The growth in domain size and towards purer structures has also been observed with resonant soft x-ray scattering.