Six models with differing representation of the physical process in the coupled soil- plant system are tested to simultaneously reproduce the dynamics of soil water contents, evapotranspiration, and leaf area index during a growing season of winter wheat at two contrasting field plots in the Kraichgau and the Swabian Alb regions in South-West Germany. The main aim of the study is the assessment of the performance and the identification of structural deficits of the models LEACHN, SUCROS, CERES, GECROS, and SPASS as well as the land-surface model CLM3.5. The calibration of each model is posed in a multiobjective framework with three different objective functions that summarize the fit between model simulations and the three observation types. The AMALGAM evolutionary search algorithm is utilized to simultaneously estimate the most important soil hydraulic and plant parameters. The six models exhibit a wide variability in the trade-offs between the fitting to the data types. Mechanistic process description, particularly of the root system, reduces the trade-off considerably for the SPASS and GECROS models. These models adequately simulate the reduction of root water uptake and transpiration during senescence under nonlimited soil water supply. The SPASS model in particular shows an overall better performance as compared to the more simpler models which is related to an adequate level of structural complexity in the interplay of all model compartments combined with a relatively low parameter sensitivity to the weighting scheme in the multiobjective optimization. The dynamic consideration of the root system formation is particularly important, which is simulated quite detailed in the SPASS model as a function of nitrogen (N) and water availability in the different soil horizons. The proposed multiobjective calibration procedure proved to be very useful to identify processes that are important to adequately simulate the coupled soil-plant system. The consideration of these processes and our insights about the value of different data types for model calibration is expected to lead to more accurate, predictive land-surface models.