The striatum, a subcortical structure, is the principal target of the neurodegenerative process in Huntington's disease (HD). The measurement of striatal atrophy using the bicaudate ratio on CT scanner images has therefore been used for years to assess disease progression, but this measure only takes into account unidimensional changes in the head of the caudate nucleus. Recently, voxel-based morphometry (VBM), which permits automated statistical comparisons of whole-brain MRI images, has been proposed to quantify striatal atrophy. However, VBM was not originally designed to study subcortical structures, and severe deep brain deformations that occur in HD may hamper the automatic processing of VBM. Here, we validate the use of the optimised protocol of VBM to quantify subcortical atrophy in HD by comparing results obtained with this method to those provided by manual segmentation of subcortical structures. We studied 20 patients with early HD and 12 controls matched for age, sex and handedness using an improved T1-weighted sequence that eased grey matter segmentation. Both manual and automated methods evidenced the dorso-ventral gradient of striatal atrophy, a loss of grey matter in the globus pallidus and the thalamus, and similar correlations between clinical scores and subcortical atrophy. Furthermore, we were able to detect with VBM grey matter loss in the substantia nigra, the hypothalamus, the amygdala, the insular cortex and the premotor and sensorimotor cortices. Finally, VBM provided results consistent with previous post mortem results and proved to be a sensitive biomarker capable of correctly managing subcortical distortions throughout HD patients' brains.