Discontinuities in flood frequency curves, here referred to as step changes, hinder the estimation of high return levels of streamflow. In this paper, we develop a robust and objective methodology for the detection of step changes, and apply it to a large dataset of catchments in the USA and Germany. Given the statistical uncertainty of observed time series due to their limited sample size, we then assess the reliability of a PHysically-based Extreme Value (PHEV) distribution of river flows to identify catchments that might experience a step change. Results show that PHEV is suitable for step changes detection, with a high correct detection rate especially in the autumn and summer seasons, whereas it tends to often show a step change not visible in the observations in spring and winter (seasons typically characterized by persistent flow regimes with reduced likelihood of exhibiting relatively large floods), for which we examine the possible reasons. By means of a controlled experiment we re-evaluate the step change detection method on true positive cases (i.e., when both observations and PHEV display a step change) discarding the highest maxima. PHEV confirms its capability to detect a step change, as observed in the original flood frequency curve, even if the shortened one does not show it. These findings prove the reliability of PHEV for the identification of step changes, especially relevant in scarce data regions, and set the premises for a deeper investigation of physiographic and hydroclimatic attributes controlling the emergence of discontinuities in flood frequency curves.