As diffusion tractography is increasingly used to generate quantitative measures to address clinical questions, it is important to characterise the inter-session reproducibility and inter-subject variability of these measures. Here, we assess the reproducibility and variability of diffusion tractography measures using diffusion data from 8 subjects scanned 3 times. We used probabilistic tractography to define the cingulum bundle, pyramidal tracts, optic radiations and genu of the corpus callosum in each individual data set using three different methods of seed definition. Measures of mean fractional anisotropy (FA) and mean diffusivity (MD) along the tracts were more reproducible than measures of tract volume. Further, tracts defined using a two region of interest (ROI) approach were more reproducible than those defined using manually placed seed masks alone. For mean FA taken from tracts defined using the two ROI approach, inter-session coefficients of variation (CV) were all below 5% and inter-subject CVs were below 10%; for mean MD inter-session, CVs were all below 3% and inter-subject CVs were below 8%. We use the variability measures found here to calculate the sample sizes required to detect changes in FA, MD or tract volume of a given size, either between groups of subjects or within subjects over time. Finally, we compare tractography results using 60 diffusion encoding directions to those found using a subset of 12 directions; the number of diffusion directions did not have a significant effect on reproducibility, but tracts derived using fewer directions were consistently smaller than those derived using 60 direction data. We suggest that 12 direction data are sufficient for reproducibly defining the core of large bundles but may be less sensitive to smaller pathways.