This paper presents a data classification algorithm able to detect corrupted measurements as outliers, with application to underwater ultra-short baseline (USBL) acoustic positioning systems. The devised framework is based on causal median filters that are readily implementable, and a set of theoretical analysis tools that allows for the design of the filter parameters is also presented. The design takes into account very specific implementation details of USBL acoustic positioning systems and also inherent non-ideal characteristics that include long period data outages. The outlier classifier is evaluated both in simulation and with experimental data from a prototype USBL acoustic positioning system fully developed in-house. © 2015 Springer International Publishing.