Fault surface roughness is a principal factor influencing earthquake mechanics, and particularly rupture initiation, propagation, and arrest, however little data currently exists on fault surfaces at seismogenic depths. Here we investigate the roughness of slip surfaces from the seismogenic strike-slip Gole Larghe Fault Zone, exhumed from ca. 10 km depth. The fault exploited pre-existing joints and is hosted in granitoid rocks of the Adamello batholith (Italian Alps). Individual seismogenic slip surfaces generally show a first phase of cataclasite production, and a second phase with beautifully preserved pseudotachylytes of variable thickness. We determined the geometry of fault traces over five orders of magnitude using terrestrial laser-scanning (LIDAR, ca. 500 m to 1 m scale), and 3D mosaics of high-resolution rectified digital photographs (10 m to <1 mm scale). LIDAR scans and photomosaics were georeferenced in 3D using a Differential Global Positioning System, allowing detailed multiscale reconstruction of fault traces. The combination of LIDAR and high-resolution photos used in this study has the advantage, over classical LIDAR-only surveys, that the spatial resolution of rectified photographs can be very high (0.2 mm/pixel in this study), allowing for detailed outcrop characterization. Fourier power spectrum analysis of the fault profiles revealed a self-affine behaviour over 3 to 5 orders of magnitude, with Hurst exponents ranging between 0.6 and 0.8. Roughness anisotropy is always small to negligible for the Gole Larghe Fault Zone, while roughness of pre-existing joints is not significantly different from the seismogenic fault surfaces. These observations are consistent with the generally small offsets shown by individual seismogenic fault surfaces, and indicate that precursor joints have a strong influence on the roughness of the fault surfaces. From a methodological point of view, the technique used here is advantageous over direct measurements of exposed fault surfaces in that it preserves, in cross-section, all of the structures which contribute to fault roughness, and removes any subjectivity introduced by the need to distinguish roughness of original slip surfaces from roughness induced by secondary weathering processes. Moreover, offsets can be measured by means of suitable markers and fault rocks are preserved, hence their thickness, composition and structural features can be characterised, providing an integrated dataset which sheds new light on mechanisms of roughness evolution with slip and concomitant fault rock production. Parameters from the Fourier analysis have been used to reconstruct synthetic 3D fault surfaces with an equivalent roughness by means of 2D Fourier synthesis, which can be in turn be used as input data for different fault numerical models.