The paper proposes an adaptation of the original median test for detecting spurious data in Particle Image Velocimetry (PIV) data. The proposed method normalizes the median residual using a robust estimate of the local velocity variation, specifically the root-mean-square (RMS) velocity fluctuation \( u' \). This normalization addresses the issue of a single threshold value being insufficient for detecting outliers in heterogeneous flow data, where the mean velocity can vary significantly. The normalized median test yields residuals that follow a more universal probability density function, allowing for a single threshold to effectively detect spurious vectors across a wide range of Reynolds numbers (from \( 10^{-1} \) to \( 10^7 \)). The method is demonstrated to be effective in various flow cases, including turbulent jets and grid-generated turbulence, by reducing the correlation between the mean residual and the turbulence level. A minimum normalization level \( \varepsilon \) is introduced to handle cases with very low turbulence intensities, further improving the robustness of the method.The paper proposes an adaptation of the original median test for detecting spurious data in Particle Image Velocimetry (PIV) data. The proposed method normalizes the median residual using a robust estimate of the local velocity variation, specifically the root-mean-square (RMS) velocity fluctuation \( u' \). This normalization addresses the issue of a single threshold value being insufficient for detecting outliers in heterogeneous flow data, where the mean velocity can vary significantly. The normalized median test yields residuals that follow a more universal probability density function, allowing for a single threshold to effectively detect spurious vectors across a wide range of Reynolds numbers (from \( 10^{-1} \) to \( 10^7 \)). The method is demonstrated to be effective in various flow cases, including turbulent jets and grid-generated turbulence, by reducing the correlation between the mean residual and the turbulence level. A minimum normalization level \( \varepsilon \) is introduced to handle cases with very low turbulence intensities, further improving the robustness of the method.