The difﬁculty in modelling inﬂation and the signiﬁcance in discovering the underlying data-generating process of in-ﬂation is expressed in an extensive literature regarding inﬂation forecasting. In this paper we evaluate nonlinearmachine learning and econometric methodologies in forecasting US inﬂation based on autoregressive and structuralmodels of the term structure. We employ two nonlinear methodologies: the econometric least absolute shrinkageand selection operator (LASSO) and the machine-learning support vector regression (SVR) method. The SVR hasnever been used before in inﬂation forecasting considering the term spread as a regressor. In doing so, we use a longmonthly dataset spanning the period 1871:1–2015:3 that covers the entire history of inﬂation in the US economy. Forcomparison purposes we also use ordinary least squares regression models as a benchmark. In order to evaluate thecontribution of the term spread in inﬂation forecasting in different time periods, we measure the out-of-sample fore-casting performance of all models using rolling window regressions. Considering various forecasting horizons, theempirical evidence suggests that the structural models do not outperform the autoregressive ones, regardless of themodel’s method. Thus we conclude that the term spread models are not more accurate than autoregressive modelsin inﬂation forecasting.