Subduction zones are naturally complex systems, with much of their deformation being accommodated along the interface between two tectonic plates. Hence, the physical nature and the rheology of the subduction interface play an important role in the deformation, degree of locking, and slip processes during convergence, as well as large scale subduction dynamics.Over the last two decades, different slip patterns have been recognized by geodetic and seismic techniques, such as slow slip events and episodic tremor and slip, the physics behind which still evade us. Since direct observations along an active subduction interface are not feasible, field examples of exhumed zones can provide important insight into the nature of such transient events. Concurrently, most field outcrops suggest that the subduction interface is rather heterogeneous, comprising different units that deform following various patterns. However, these units are often too small to be resolved in a large-scale geodynamic model and are, therefore,not taken into account in such models. Outcrop-scale numerical experiments can be combined with natural structural observations from outcrops in order to refine the rheologies used in large geodynamic models. Here, methods ranging from field, petrological and geochronological observations on exhumed rocks to outcrop-scale numerical simulations are deployed, in order to investigate the rheology of the plate interface, the deformation mechanisms and the timing of deformation. The European Alps are a great natural laboratory exposing an almost continuous subduction interface allowing for the study of deformation processes from shallow to deeper segments (from ca. 10 km to ca. 45 km depth). Here, a network of fossil subduction plate interfaces preserved in the Central Alps (Val Malenco, N Italy) is used as a proxy to study such processes on subducting continental slices (the Margna and Sella nappes), at depths corresponding to the former brittle-ductile transition. This network of shear zones comprises mostly mylonites and schists but also rare foliated cataclasites, with different generations of micas and garnet locally overgrowing resorbed pre-Alpine cores. Thermodynamic modelling points to peak burial deformation conditions of ~0.9 GPa and 350°-400° C, at ca. 30 km depth. Rb/Sr geochronology on marbles yields an age of 48.9 ± 0.9 Ma, while a wide range of both Rb/Sr and 40Ar/39Ar apparent ages is obtained from deformed orthogneisses and micaschists embracing 87-44 Ma, due to incomplete recrystallization. Based on pressure-temperature, structural and geochronological observations, the studied shear zones last equilibrated at depths downdip of the seismogenic zone in an active subduction zone setting. Fluids contribute to the bulk rheology of this interface by enhancing pressure-solution creep which prevails in the microstructural record. This study suggests that this system of shear zones represents deformation conditions along the subduction interface(s) in the transition zone below the seismogenic zone during active subduction, where transient slip is found. Other exhumed subduction interfaces exhibit block-in-matrix characteristics, termed mélanges, the block concentration of which can affect their bulk rheology. To investigate this, synthetic models are created, with different proportions of strong blocks within a weak matrix, and compared to exhumed natural mélanges outcrops. 2D Finite Element visco-plastic models in simple shear are used to determine the effective rheological parameters of such a two-phase medium, comprising blocks of basalt within a wet quartzitic matrix. Models and their structures are treated as scale-independent and self-similar. Therefore, field geometries are upscaled to km-scale models, compatible with large-scale, geodetic and seismic observations. Outcrops of mélanges, as well as of other units deformed during subduction suggest that deformation is mainly taken up by dissolution-precipitation creep. However, flow laws for dissolution-precipitation creep are not well-established experimentally and scarcely used in large-scale numerical models. To make the results of this study comparable to and usable by numerical studies, dislocation creep is assumed to be the governing flow law for both phases (basalt and wet quartz). Finally, effective rheological estimates for a natural subduction interface are provided. The results suggest that block concentration affects deformation and strain localization, with the effective dislocation creep parameters (A, n, and Q) varying between the values of the strong and the weak phase, when both phases deform viscously. However, as the contribution of brittle deformation of the basaltic blocks increases, the value of the stress exponent, n, can exceed that of the purely strong phase. Using these effective parameters as input into seismic cycle models could help evaluate the possible effect of field heterogeneities on the slip behaviour of the plate interface. In summary, the heterogeneity of the subduction interface plays an important role in the degree of localization and rheology of the plate interface. Mixed brittle-ductile deformation is common in subducted rocks and might give rise to different kinematic behaviours. Re-assessing fabrics in exhumed rocks with respect to their (relevant) timing, spatial distribution, as well as cross-cutting relationships of individual fabric features is essential for linking kinematic far-field observations to the physics of deformation processes acting upon the interface. Finally, incorporating the results of small-scale numerical studies in large-scale geodynamic models may help improve our understanding of the mechanical behaviour of the plate interface, including transient or aseismic slip phenomena, which may control the recurrence of megathrust ruptures in active subduction systems.