Since laboratory experiments point to the existence of both diffusion creep and power-law creep at realistic mantle conditions, a composite rheology in which both mechanisms operate at the same time so that their strains are combined might be more realistic and has in the past been shown to provide a better fit to observations of glacial isostatic adjustment (GIA) than purely linear rheology ( Gasperini et al., 2004; Dal Forno et al., 2005). To further investigate the effect of such rheology on sea level curves and uplift rate resulting from GIA, composite rheology is implemented in the coupled Laplace-finite element method for a 3D spherical self-gravitating Earth. We vary the pre-stress exponent (assumed to be derived from a uni-axial stress experiment) and the Newtonian viscosity for a homogeneous mantle. Composite rheology is found to have a statistically significant better fit with observed relative sea level data than linear rheology (diffusion creep only) and non-linear rheology (dislocation creep only). For the best-fitting composite rheology model it is shown that in the mantle below the former ice sheet margin, stress is high enough for power-law creep to become dominant during melting and shortly thereafter, causing the model to behave mostly in a non-linear way. It is found that composite rheology, with the parameters investigated in this paper, not only provides a better fit to sea level data than non-linear rheology but also slightly increases present-day uplift rate compared to non-linear rheology. This encourages application of composite rheology in GIA models that aim to improve knowledge of mantle rheology. Low uplift rates for composite rheology can be further increased by a large increase in ice thickness in North America at the expense of violating total melt-water constraints. A 1 or 2 ka delay in Laurentide glaciation and deglaciation increases uplift rates for all values of the pre-stress exponent investigated, while fit to a number of relative sea level observations in the Laurentide ice sheet is improved. Large increase in ice thickness disagrees with other observations (total melt constraints), therefore a delay in glaciation is a promising direction if global ice models are to be adjusted for a composite rheology.