The viscosity of the mantle is indispensable for predicting Earth's mechanical behavior at scales ranging from deep mantle material flow to local stress accumulation in earthquakes zones. But, mantle viscosity is not well determined. For the lower mantle, particularly, only few constraints result from elaborate high-pressure experiments (Karato, 2008) and a variety of viscosity depth profiles result from joint inversion of the geoid and postglacial rebound data (Forte and Mitrovica, 1996; Kaufmann and Lambeck, 2000; Mitrovica and Forte, 2004). Here, we use inferred lower-mantle sinking speed of lithosphere subduction remnants as a unique internal constraint on modeling the viscosity profile. This entails a series of elaborate dynamic subduction calculations spanning a range of viscosity profiles from which we select profiles that predict the inferred sinking speed of 12±3mm/yr (van der Meer et al., 2010). Our modeling shows that sinking speed is very sensitive to lower mantle viscosity. Good predictions of sinking speed are obtained for nearly constant lower mantle viscosity of about 3-4×10 22Pas. Viscosity profiles incorporating a viscosity maximum in the deep lower mantle, as proposed in numerous studies, only lead to a good prediction in combination with a weak postperovskite layer at the bottom of the lower mantle, and only for a depth average viscosity of 5×10 22Pas.