Cross-flow microfiltration has replaced conventional fining methods for clarifying fruit juices, resulting in the combination of clarification and preservation in one single continuous operation. However the performance of this process is limited by the fouling mechanism during the filtration operation, related to the fruit juice fouling propensity. The aim of this work was to study the possibility of predicting the fruit juices filterability, according to some of their intrinsic characteristics. Therefore, 9 different commercial orange juices were chemically and physicochemically characterized (17 characterization parameters) and laboratory scale filtration tests, in an agitated and pressurized filtration module, were carried out at constant transmembrane pressure (1.5 bar). Fruit juices filterability was evaluated from the permeate flux evolution during the filtration operation. Partial Least Squares (PLS) method was particularly used for the analysis of relationship between the fruit juices filterability and the juice studied characteristics. PLS was initially performed using the whole set of predictor variables. The aim of this first analysis was to highlight the most important juice characteristics among the 17 studied ones; a simplified PLS model was performed using model-wise method and PLS-VIP method as important variable selection procedure. It was possible to predict satisfactorily (R2 > 0.9) fruit juices filterability according to only 5 physicochemical characteristics, reflecting the different phases of the studied juices (dry matter, total soluble solids, pH, conductivity, and surface area average diameter). (Résumé d'auteur)