The ability to infer axonal conduction velocities (CV) non-invasively from in vivo neuroimaging is of huge biological importance. Having previously shown that accurate CV estimates are feasible with MRI-measurable parameters, we show here the sensitivity of MRI-derived CV estimates to modelling errors and acquisition noise. We show that for the parameters typically seen in white matter axons, there is less than 5% error in CV estimates. Application to a human diffusion/relaxometry dataset generates CV estimates in corpus callosum that are close to those observed in electrophysiology literature. This illustrates further the feasibility of estimating CV from in vivo microstructural MRI.