Diffusion of scandium (Sc) in synthetic forsterite (Mg2SiO4) was studied at 1400 °C at atmospheric pressure, in air. The Sc source was a powder of forsterite and protoenstatite (Mg2Si2O6) in order to buffer the silica activity, doped with various Sc contents, from ∼9 wt. ppm Sc to ∼8.5 wt% Sc2O3. The powder with the highest Sc content also contained thortveitite (Sc2Si2O7), which additionally buffers the activity of Sc2O3. All experiments were conducted for the same duration (550 h), but the length of the Sc diffusion profile (i.e. the distance over which the concentration decreased from the rim to core/background value) varied from ∼200 to 800 μm, with higher Sc concentrations associated with longer diffusion profiles. The geometry of Sc diffusion profiles also changes systematically with the Sc concentration – profiles with low Sc (<10 wt. ppm at the interface) have stepped shapes, those with medium Sc contents (80–400 wt. ppm at the interface) have near-linear profiles, while experiments with high Sc at the interface (>500 wt. ppm) show profiles with a shape that is approximately, but not exactly, describable by error function forms. Small but variable amounts of Fe contamination resulted in Fe diffusion profiles with interface concentrations of 80 to 1100 wt. ppm. The diffusivity of Fe, at these trace concentrations, appears to be a function of the concentration of Sc, rather than its own concentration, with the increased Sc concentration associated with an increase in Fe diffusivity of at least an order of magnitude. Similar behaviour is observed for other contaminants, Ni, Co, Mn and Li, at trace concentrations. Additionally, Al, present in the original forsterite crystal at the level of a few 10s of wt. ppm developed concentration profiles with a stepped form, broadly showing out-diffusion. The point along the profiles at which the Al concentration changes dramatically, coincides approximately with the end of the Sc profiles. Models are tested in which Sc diffusion is assumed to be concentration-dependent. In these models, the concentration-dependence is defined using the energy of binding between Sc3+ and M-site vacancy, and the diffusivity of the cation-vacancy pair. Such models are generally successful, but cannot fully describe the behaviour observed in these experiments. Rather, the diffusive behaviour of Sc and Al is better described by a model in which Sc3+ substitutes for Mg2+, charge balanced by vacancies (the Sc4/3[vac]2/3SiO4 substitution), and diffuses relatively rapidly in a concentration-dependent regime. Interaction of Sc with Al (present in the starting material) is proposed to lead to the formation of an MgScAlO4 component, upon which the diffusivity of Sc is drastically reduced. The model is capable of describing the majority of Sc and Al profiles observed. One experiment was hydroxylated following the diffusion anneal, revealing a multitude of defects supporting the suggestion that this system is more complex than simple concentration-dependent diffusion. Whilst these experiments were conducted in a simple system with lower Fe concentrations and higher fO2 than those typical of natural olivine, an example from a previous study (Cr diffusion in San Carlos olivine) shows that such complex behaviour may also be found in systems closer to natural conditions. The addition of Cr (∼2000 wt. ppm at the diffusion interface) appears to cause a ∼ 1 order of magnitude increase in the diffusivity of Fe-Mg, Ca and Ni relative to the published values conventionally used for diffusion chronometry.