Ideally, animal models of neurodegenerative diseases should reproduce the clinical manifestation of the disease and a selective neuronal loss. In this review we will take as an example Parkinson's disease because its pathophysiology is well known and the neuronal loss well characterized. Indeed, Parkinson's disease is characterized by a loss of some but not all dopaminergic neurons, a loss of some non dopaminergic neurons and alpha-synuclein positive inclusions resembling Lewy bodies. There are at least two ways to develop animal models of PD based on the etiology of the disease and consist in 1) reproducing in animals the mutations seen in inherited forms of PD; 2) intoxicating animals with putative environmental toxins causing PD. In this review we discuss the advantages and the drawbacks in term of neuroproction of the currently used models.