Motion during time-series data acquisition causes model-fitting errors in quantitative dynamic contrast-enhanced (DCE) MRI studies. Motion correction techniques using conventional registration cost functions may produce biased results because they were not designed to deal with the time-varying information content due to contrast enhancement. We present a locally-controlled, 3D translational registration process driven by tracer kinetic modeling that successfully registers abdominal DCE-MRI data at high temporal resolution and compare this method to a similar approach based on registration to the time series mean image in data from 8 patients. When the registration is driven by an appropriate model, we find significant improvements in model-fitting. Also, model-driven registration influences parameter estimates and reduces repeat study variability in measurements of blood volume.