Money-demand specifications exhibit instability, especially for long spans of data. This paper reconsiders the welfare cost of inflation for the US economy using a flexible time-varying (TV) cointegration methodology to estimate the money-demand function. We find evidence that the TV cointegration estimation provides a better fit of the actual data than a time-invariant estimation and that the throughout unitary income elasticity only exists for the log–log form over the entire sample period. Our estimate of the welfare cost of inflation for a 10% inflation rate lies in the range of 0.025–0.75% of gross domestic product (GDP) and averages 0.27%. In sum, our findings fall well within the ranges of existing studies of the welfare cost of inflation. We find that the welfare cost averages 7.4% higher during expansions than recessions for 10% inflation rate. Finally, the interest elasticity of money demand shows substantial variability over our sample period.