Inflation is a far from homogeneous phenomenon, a fact often neglected in modeling consumer price inflation. Using a novel methodology grounded in theory, the ten sub-components of the consumer price index (excluding mortgage interest rates), are modeled separately and forecast, four-quartersahead. Equilibrium correction models in a rich multivariate form employ general and sectoral information, and take account of structural breaks and institutional changes. Our methods allow for longer lags than conventionally considered in VARs, but in a parsimonious manner. Sign priors are imposed on long-run effects and automatic model selection is used to select parsimonious models from more general ones. The models throw light on sectoral sources of inflation, useful to monetary policy. Data for 1979 to 2003 are used for model selection, and pseudo out of sample forecasting performance to the end of 2007 is examined. Aggregating the weighted sub-component forecasts indicates gains are made over forecasting the overall index using these methods, and also substantial gains over forecasting using benchmark naïve models. To extend this work, including sectoral information such as an explicit treatment of tax policy, regulatory information and announced administered price rises, should further enhance these forecasting methods.