Water infiltration into soil plays an important role in runoff generation, soil erosion, and irrigation planning. Accurate estimation of infiltration is essential for effective soil and water management. This study investigated temporal variations in water infiltration and evaluated the performance of four infiltration models, namely Philip, Kostiakov, Kostiakov–Lewis, and Horton, in eight watersheds of Ardabil Province, Iran, including Aladizgeh, Iril, Sula, ShamsAbad, Amuqin, Namin, Nouran, and Nir. Infiltration experiments were conducted using a double-ring infiltrometer. Model performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and Nash–Sutcliffe efficiency (NSE). Relationships between model parameters and soil properties were examined using Pearson correlation analysis. The results showed that the Philip model performed best in six watersheds (Aladizgeh, Iril, Sula, Amuqin, Namin, and Nouran), while the Kostiakov and Kostiakov–Lewis models performed best in ShamsAbad and Nir, respectively. Overall, the Philip model was the most suitable model, with mean R² and NSE values of 0.975 and 0.927, respectively. Correlation analysis indicated that the Philip capillary sorptivity coefficient (S) was significantly positively correlated with clay content (r = 0.82) and negatively correlated with sand content (r = −0.76) at the 5% significance level. These findings demonstrate that infiltration model performance depends on local soil conditions and that model selection should consider soil properties and multiple performance criteria. The Philip model is therefore recommended for most semi-arid watersheds in Ardabil Province.