Deep-seated metamorphic rocks are commonly found in the interior of many divergent and convergent orogens. Plate tectonics can account for high-pressure metamorphism by subduction and crustal thickening, but the return of these metamorphosed crustal rocks back to the surface is a more complicated problem. In particular, we seek to know how various processes, such as normal faulting, ductile thinning, and erosion, contribute to the exhumation of metamorphic rocks, and what evidence can be used to distinguish between these different exhumation processes.In this paper, we provide a selective overview of the issues associated with the exhumation problem. We start with a discussion of the terms exhumation, denudation and erosion, and follow with a summary of relevant tectonic parameters. Then, we review the characteristics of exhumation in different tectonic settings. For instance, continental rifts, such as the severely extended Basin-and-Range province, appear to exhume only middle and upper crustal rocks, whereas continental collision zones expose rocks from 125 km and greater. Mantle rocks are locally exhumed in oceanic rifts and transform zones, probably due to the relatively thin crust associated with oceanic lithosphere.Another topic is the use of P-T-t data to distinguish between different exhumation processes. We conclude that this approach is generally not very diagnostic since erosion and normal faulting show the same range of exhumation rates, reaching maximum rates of >5-10 km Ma -1 for both processes. In contrast, ductile thinning appears to operate at significantly slower rates. The pattern of cooling ages can be used to distinguish between different exhumation processes. Normal faulting generally shows an asymmetric distribution of cooling ages, with an abrupt discontinuity at the causative fault, whereas erosional exhumation is typically characterized by a smoothly varying cooling-age pattern with few to no structural breaks. Last, we consider the challenging problem of ultrahigh-pressure crustal rocks, which indicate metamorphism at depths greater than 100-125 kin. Understanding the exhumation of these rocks requires that we first know where and how they were formed. One explanation is that metamorphism occurred within a thickened crustal root, but it does seem unlikely that the crust, including an eclogitized mafic lower crust, could get much thicker than c. 110 km while maintaining a reasonable Moho depth (<70 km, assuming that the seismically defined Moho would be observed to lie above the eclogitized lower crust). Diamondbearing crustal rocks cannot be explained by this scheme. The alternative is to accrete the upper 10-40 km of lithospheric mantle into the orogenic root. This scenario will provide sufficient pressures for both coesite-and diamond-bearing eclogite-facies metamorphism, while maintaining a reasonable Moho depth (<70 km) and reasonable mean topography (<3 kin). We speculate that the detachment and foundering of the mantle root may contribute to the exhumation of any crust...