This paper presents a condition‐based treatmentmethodology for a type ofrailsurface defect called “squat”. The proposed methodology is based on a set of robust and predictive fuzzy key performance indicators. A fuzzy Takagi Sugeno interval model is used to predict squat evolution for different scenarios over a time horizon. Models including the effects of maintenance to treat squats, via either grinding or replacement of the rail, are also described. A railway track may contain a huge number of squats distributed in the rail surface with different levels ofseverity. We propose to aggregate the localsquat interval modelsinto track‐level performance indicatorsincluding the number and density ofsquats per track partition. To facilitate the analysis of the overall condition, we propose a single fuzzy global performance indicator per track partition based on a fuzzy expert system that combines all the scenarios and predictions over time. The proposed methodology relies on the early detection ofsquats using Axle Box Acceleration measurements. We use real‐life measurementsfrom the track Meppel‐Leeuwarden in the Dutch railway network to show the benefits of the proposed methodology. The use of robust and predictive fuzzy performance indicators facilitates the visualization ofthe track health condition and easesthemaintenance decision process.