We examine the inflation-output relationship in the USA for the period 1955-90. We start by replicating Smyth (1992) and subjecting his estimates to a series of diagnostic tests. The model is shown to satisfy conditions for valid inference (weak exogeneity) and policy analysis (super-exogeneity) (Engle, Hendry and Richard, 1983). These robustness checks allow us to study the out-of-sample consequences of the point estimates for various levels of inflation. One central experimental result is that for inflation rates exceeding 4% the natural rate of output is reduced to such an extent that it contributes to a reduction in the growth rate of real GNP that is below historical trend (3.1% in our sample).