Docking is the computational prediction of the 3D structure of a molecular complex, and a powerful tool for studying them. The high flexibility of molecules is a scourge of docking, as it leads to a very large number of molecular conformations, and thereof of putative docking models, processing of which is computationally expensive. In addition, it can be very challenging to separate correct docking models from incorrect ones. The described issue can be encountered in the scoring procedure of a fragment-based docking [1] of single-stranded RNAs (highly flexible molecules) to proteins. During this procedure, each generated docking model is given a score which is a sum of atom-atom distance-based pseudo-energy of interaction. Models with the lowest score are considered to be correct. Often incorrect models get a low score, while truly correct ones get a high score. This issue can be addressed by introducing a new way to score the models. Here we are proposing to use atom-atom distances in the modelled ssRNA-protein interfaces for the identification of the correct models. For each pair of atom types present in a docking model, we create a histogram describing the log-odd (a way to describe probability) of a given atom-atom distance to occur in a correct model. With the set of histograms for all pairs of atom types, we can give a histogram-based score to a docking model by summing log-odds of all atom-atom distances present in a model. Models with high histogram scores are likely to be correct. We tested our approach on a small benchmark of 131 complexes of known structures of ssRNA trinucleotides bound to 56 proteins. We derived 2 independent sets of histograms from docking models of 2 randomly selected complexes and used them to score the rest of the complexes (so each model had 3 scores: default one, histogram-based by the first set of histograms, and histogram-based by the second set of histograms). To evaluate if, for most complexes, at least one of the histogram sets is appropriate to score the models, we pool together the top 10% ranked models by each set. We then compare what percentage of the correct models ends up in that pool versus in the top 20% ranked models by the default score. Histograms percentage was higher than the default for 49% of the complexes, similar for 27% of the complexes and lower for 24% of the complexes. In perspective, if we can find a way to transform a set of histograms into a set of docking parameters then we can use this approach to increase the raw number of correct models generated during the docking thus increasing the reliability of the docking. 1. Chauvot de Beauchene I, de Vries SJ, Zacharias M., Fragment-based modelling of single stranded RNA bound to RNA recognition motif containing proteins, Nucleic Acids Research Vol. 44, No. 10, 4565–4580 (2016)