Identifying biomarkers which can be used as a diagnostic tool is a major objective of pharmacogenetic studies. Most mental and many neurological disorders have a compiled multifaceted nature, which may be the reason why this endeavor has hitherto not been very successful. This is also true for tardive dyskinesia (TD), an involuntary movement complication of long-term treatment with antipsychotic drugs. The observed associations of specific gene variants with the prevalence and severity of a disorder can also be applied to try to elucidate the pathogenesis of the condition. In this paper, this strategy is used by combining pharmacogenetic knowledge with theories on the possible role of a dysfunction of specific cellular elements of neostriatal parts of the (dorsal) extrapyramidal circuits: various glutamatergic terminals, medium spiny neurons, striatal interneurons and ascending monoaminergic fibers. A peculiar finding is that genetic variants which would be expected to increase the neostriatal dopamine concentration are not associated with the prevalence and severity of TD. Moreover, modifying the sensitivity to glutamatergic long-term potentiation (and excitotoxicity) shows a relationship with levodopa-induced dyskinesia, but not with TD. Contrasting this, TD is associated with genetic variants that modify vulnerability to oxidative stress. Reducing the oxidative stress burden of medium spiny neurons may also be the mechanism behind the protective influence of 5-HT2 receptor antagonists. It is probably worthwhile to discriminate between neostriatal matrix and striosomal compartments when studying the mechanism of TD and between orofacial and limb-truncal components in epidemiological studies.