We provide a new experimental investigation of the neutrality theorem of Warr (1983), according to which "when a single public good is provided at positive levels by private individuals, its provision is unaffected by a redistribution of income". Our experimental design differs from Chan et al. (1996) in that we redistribute within each group the total group endowment after 10 rounds, instead of comparing different income distributions across groups. We need therefore to control for the "restart effect" (Andreoni, 1988). To do so, we compare our two test-treatments, an unequalizing redistribution (EI) and an equalizing redistribution (IE), to two benchmark treatments for which the same sequence of 10 rounds is repeated twice either with an equal distribution (EE) or with unequal distribution (II). The constituent game has a unique interior dominant strategy equilibrium depending on income. Our data supports the neutrality theorem (after controlling for the restart effect) : redistribution does not affect the total amount of public good in none of the test treatments. However, the analysis of individual behavior shows that "poor" subjects over-contribute with respect to their Nash contribution, will "rich" subjects tend to play their Nash-contribution or under-contribute slightly. Furthermore, after a redistribution, both types of subjects tend to under-react : those who get poor do not reduce their contribution enough, while those who get richer do not increase sufficiently their contribution.