Rheological structure of subducted slabs of oceanic lithosphere in the mantle transition zone is investigated based on mineral physics observations incorporating grain-size, stress, temperature and pressure dependence of rheology. It is shown that the rheological structure of slabs depends strongly on subduction parameters through temperature that controls the grain-size of spinel (ringwoodite) and the magnitude of forces acting on a slab. We use a theoretical model of grain-size evolution associated with the olivine–spinel transformation, plastic flow laws of olivine and spinel combined with a thermal model of subducting slab in which the effect of latent heat release is incorporated. Three deformation mechanisms for olivine and spinel (diffusional creep, power-law (dislocation) creep and the Peierls mechanism) are considered. Due to the large variation in temperature, stress and grain-size, a subducting slab is shown to have a complicated rheological structure which varies both laterally and with depth. A cold slab in the deep transition zone is characterized by a weak, fine-grained spinel region surrounded by narrow but strong regions. The flexural rigidity and the curvature of a slab are calculated using a new formulation in which the effects of stress-dependent rheology is incorporated in a self-consistent fashion. Although uncertainties in both the transformation kinetics and the rheology of high pressure phases are still large, the general trend of dependence of slab flexural rigidity and the curvature on the subduction parameters is well constrained. Slabs with very low thermal parameters (warm slabs) are weak, but slabs with large thermal parameters (cold slabs) are also weak due to small spinel grain-size and large external force (bending moment). Slabs with intermediate thermal parameters will have a relatively large flexural rigidity and could penetrate into the lower mantle without much deformation. Thus the 660 km discontinuity may work as a rheological filter for mantle convection. This prediction provides a natural explanation for a paradoxical observation that significant deformation of slabs is observed exclusively in the western Pacific where temperatures of the slabs are considered to be low. Our slab rheology models also have important implications for deep earthquakes. Overall rheological weakening of slabs in the deep transition zone results in high rates of deformation under relatively low temperatures providing a favorable environment for thermal runaway instability (adiabatic shear instability). Our model predicts heterogeneous energy dissipation as a result of heterogeneous rheology: energy dissipation in deep, cold slabs is concentrated in high strength regions surrounding a weak, fine-grained spinel core. The regions of high energy dissipation are prone to thermal runaway instability and are likely to be seismogenic. The width of this seismogenic region is well constrained by our model and is predicted to be ∼40–60 km which is in excellent agreement with seismological observations. Other features of deep earthquakes including low seismic efficiency and low aftershock activities can also be explained by the thermal instability model.