Reliability of perfusion parameters is an essential condition to base therapy monitoring on their evolution. To test this reliability, CEUS imaging with controlled injection of a bolus of micro-bubbles was performed in a test-retest study in four mice with ectopic tumors. A conventionally used explicit model, the log-normal model (LN), a compartmental model based on an arterial input function (AIF), and a compartmental model using the kinetics in a reference tissue (RT) were compared in regards to the within-animal coefficients of variation of the volume and flow parameters estimated in tumor. The LN model proved to be the less reproducible while the RT model was the most robust. LN model was significantly outperformed by both AIF and RT models for both flow and volume parameters. This study underlines the practical interest of the RT approach for a robust inter-study comparison of perfusion parameters using CEUS imaging.