The impact of feature normalization methods on radiomic models was investigated. The study compared seven normalization methods—z-Score, robust z-Score, Min-Max, power, quantile, and tanh transformations—on fifteen publicly available radiomic datasets. The goal was to assess their effect on predictive performance, feature selection, and model calibration. The results showed that the differences in predictive performance between normalization methods were generally small, with the largest gain in AUC being +0.051 for the robust z-Score (5,95) method. The z-Score performed best overall, while the tanh transformation performed the worst. Feature selection varied significantly across methods, with only mild agreement between selected features. Model calibration was not significantly affected by normalization methods. Applying normalization before cross-validation did not introduce significant bias. The choice of normalization method influenced predictive performance and feature selection, but its impact depended on the dataset. The study emphasizes the importance of feature normalization in radiomics, as it affects both model performance and feature interpretation. It is recommended to test multiple normalization methods to achieve the best predictive performance. The study also highlights the need for further research on the impact of normalization methods in high-dimensional radiomic datasets.The impact of feature normalization methods on radiomic models was investigated. The study compared seven normalization methods—z-Score, robust z-Score, Min-Max, power, quantile, and tanh transformations—on fifteen publicly available radiomic datasets. The goal was to assess their effect on predictive performance, feature selection, and model calibration. The results showed that the differences in predictive performance between normalization methods were generally small, with the largest gain in AUC being +0.051 for the robust z-Score (5,95) method. The z-Score performed best overall, while the tanh transformation performed the worst. Feature selection varied significantly across methods, with only mild agreement between selected features. Model calibration was not significantly affected by normalization methods. Applying normalization before cross-validation did not introduce significant bias. The choice of normalization method influenced predictive performance and feature selection, but its impact depended on the dataset. The study emphasizes the importance of feature normalization in radiomics, as it affects both model performance and feature interpretation. It is recommended to test multiple normalization methods to achieve the best predictive performance. The study also highlights the need for further research on the impact of normalization methods in high-dimensional radiomic datasets.