In agronomy and hydrology, models are used to analyze experimental data, whereas experiments are needed to parameterize models. The sensitivity of model outcomes to input parameters is a key issue in this context. As a contribution to the subject, the objective of this study was to evaluate some of the sensitivities of the agro-hydrological model SWAP for predicting the water-limited yield of a winter wheat crop using soil texture data to obtain soil hydraulic parameters by a pedotransfer function. The sensitivity of yield predictions to the overall variability of involved parameters was tested. The highest prediction accuracy obtained with the calibrated SWAP model was similar to the standard deviation and about 5% of observed yields. The drainage outflow was shown to be a determining component of the soil water balance for yield predictions in the evaluated scenario, making its correct estimation important, hydraulic conductivity being the key property determining this outflow. Calibration of a transpiration reduction function using SWAP in combination with an observed yield dataset is subject to the model sensitivity to soil hydraulic properties. Such calibration by inverse modeling is only feasible if soil hydraulic properties, including the unsaturated hydraulic conductivity function, are known with high accuracy. Sensitivity of predictions of soil water content to empirical transpiration reduction parameters is low, making their calibration a less important issue when SWAP is used to inversely model soil hydraulic properties. Yield predictions showed to be highly sensitive to soil hydraulic parameterization, even when this sensitivity does not show up strongly in predicted soil water content or pressure head values. It is emphasized that experimentation to support model and calibration improvements should include the detailed determination of soil hydraulic properties, including unsaturated hydraulic conductivity, together with observation of crop and soil parameters over time.