Many per- and polyfluoroalkyl substances (PFASs) have been identified in the environment, and some have been shown to be extremely persistent and even toxic, thus raising concerns about their effects on human health and the environment. Despite this, little is known about most PFASs. In this study, the comprehensive database of over 4700 PFAS entries recently compiled by the OECD was curated and the chemical variation was analysed in detail. The analysis revealed 3363 individual PFASs with a huge variation in chemical functionalities and a wide range of mixtures and polymers. A hierarchical clustering methodology was employed on the curated database, which resulted in 12 groups, where only half were populated by well-studied compounds thus indicating the large knowledge gaps. We selected both a theoretical and a procurable training set that covered a substantial part of the chemical domain based on these clusters. Several computational models to predict physicochemical and environmental fate related properties were assessed, which indicated their lack of applicability for PFASs and the urgent need for experimental data for training and validating these models. Our findings indicate reasonable predictions of the octanol-water partition coefficient for a small chemical domain of PFASs but large data gaps and uncertainties for water solubility, bioconcentration factor, and acid dissociation factor predictions. Improved computational tools are necessary for assessing risks of PFASs and for including suggested training set compounds in future testing of both physicochemical and effect-related data. This should provide a solid basis for better chemical understanding and future model development purposes.