There is already a small literature emphasising the empirical failure of the New Keynesian IS curve, but it is not yet known if this failure reflects empirical problems associated with small samples or is rather a structural weakness of the underlying model. To address this question, in this paper I estimate the New Keynesian IS curve for output and consumption and several possible extensions on panel data from 22 OECD countries over 40 years of data. I also evaluate whether the key parameters of the IS curve change according to countries' economic and financial structure. The main finding is that output and consumption are mainly forward looking, and this is a very robust feature of the data. At the same time, I find little evidence in favour of the traditional specification where the real interest rate enters with a negative sign due to intertemporal substitution; on the contrary, it is typically either insignificant or wrongly signed. Overall, I conclude that the New Keynesian IS curve, at least in its most common formulations, is not structural and is overwhelmingly rejected by the data.