Mass transfer in biofilms is determined by diffusion. Different mostly invasive approaches have been used to measure diffusion coefficients in biofilms, however, data on heterogeneous biomass under realistic conditions is still missing. To non-invasively elucidate fluid-structure interactions in complex multispecies biofilms pulsed field gradient-nuclear magnetic resonance (PFG-NMR) was applied to measure the water diffusion in five different types of biomass aggregates: one type of sludge flocs, two types of biofilm, and two types of granules. Data analysis is an important issue when measuring heterogeneous systems and is shown to significantly influence the interpretation and understanding of water diffusion. With respect to numerical reproducibility and physico-chemical interpretation, different data processing methods were explored: (bi)-exponential data analysis and the Γ distribution model. Furthermore, the diffusion coefficient distribution in relation to relaxation was studied by D-T2 maps obtained by 2D inverse Laplace transform (2D ILT). The results show that the effective diffusion coefficients for all biofilm samples ranged from 0.36 to 0.96 relative to that of water. NMR diffusion was linked to biofilm structure (e.g., biomass density, organic and inorganic matter) as observed by magnetic resonance imaging and to traditional biofilm parameters: diffusion was most restricted in granules with compact structures, and fast diffusion was found in heterotrophic biofilms with fluffy structures. The effective diffusion coefficients in the biomass were found to be broadly distributed because of internal biomass heterogeneities, such as gas bubbles, precipitates, and locally changing biofilm densities. Thus, estimations based on biofilm bulk properties in multispecies systems can be overestimated and mean diffusion coefficients might not be sufficiently informative to describe mass transport in biofilms and the near bulk.