Fatigue cracks in rails are submitted to mixed-mode I, II and III loadings with varying proportions of each mode along the front. In order to predict their paths and growth rates, the thresholds, bifurcation criteria, and kinetics for shear modes fatigue crack growth are needed. To obtain such data, combined mode II and III fatigue crack growth tests are run on a R260 steel, using a special asymmetric four point bending device which allows an arbitrary inclination of the pre-crack front relative to the vertical shear force, and thus a change in the proportions of mode II and III, which according to 3D finite element computations, also vary along the crack front, for a given inclination angle. This arbitrary inclination makes direct capture of images for DIC measurements of the displacements around the crack tip difficult, if not impossible. Such measurements are however necessary to derive the effective stress intensity factors, which are much lower than the nominal values, due to contact and friction of rough crack faces. The tests are thus periodically interrupted, and plastic replicas are made at minimum and maximum load in a cycle, from which positive replicas are cast, and observed with a digital optical microscope, and an AFM. The former is used to measure the in-plane relative sliding displacements of the crack faces associated with mode II, and potential asperity-induced opening displacements -usually associated with mode I- by DIC (based on a contrast introduced by sand-blasting). AFM allows very accurate measurements of the out-of-plane sliding displacements associated with mode III. The measured sliding displacement ranges are compared to those obtained by elastic-plastic F.E. computations for a smooth and frictionless crack, in order to estimate the ratio of nominal to effective DKII and DKIII. These effective SIFs are used to analyse the observed crack paths and local growth rates. The latter are deduced from post-mortem SEM and X-ray micro-tomography measurements of the initial and final positions and shapes of the crack front after a small number of cycles. Since the SIFs vary along the crack front, each test provides several data-points on the crack growth kinetics curves. The measured extent of coplanar shear-mode crack growth before bifurcation is compared to that predicted by the application, ahead of the crack front, of a shear-driven fatigue damage model identified from cyclic torsion tests on smooth specimens. Shear-mode crack growth often ends with forking in two branches at nearly ±90° to the initial crack plane -which is close to the directions of secondary maxima in the shear stress-, followed by bifurcation and mode I growth of those branches. The stress redistributions and shielding effects associated with such branching will be analyzed.