This paper presents a method for forecasting macroeconomic time series using a large number of predictors, summarized by a small number of indexes constructed via principal component analysis. The method is based on an approximate dynamic factor model, which is used to estimate the indexes and construct forecasts. The approach is applied to eight monthly U.S. macroeconomic time series, using 215 predictors in simulated real time from 1970 through 1998. The results show that the new forecasts outperform univariate autoregressions, small vector autoregressions, and leading indicator models. The method involves two steps: first, estimating the factors using principal components, and then using these factors to forecast the variable of interest. The estimated factors are referred to as diffusion indexes. The performance of the diffusion index forecasts is examined across different forecasting horizons (6, 12, and 24 months). The results indicate that the diffusion index forecasts, based on a small number of factors, perform well, with relative performance improving as the horizon increases. The improvement over the benchmark forecasts can be dramatic, in several cases producing simulated out-of-sample mean square forecast errors that are one-third less than those of the benchmark models. The paper also discusses the empirical factors and their characteristics, showing that a small number of factors can capture most of the forecasting improvement. The results suggest that a very small state vector may be necessary for forecasting macroeconomic time series. The paper concludes that the diffusion index method is a promising approach for macroeconomic forecasting.This paper presents a method for forecasting macroeconomic time series using a large number of predictors, summarized by a small number of indexes constructed via principal component analysis. The method is based on an approximate dynamic factor model, which is used to estimate the indexes and construct forecasts. The approach is applied to eight monthly U.S. macroeconomic time series, using 215 predictors in simulated real time from 1970 through 1998. The results show that the new forecasts outperform univariate autoregressions, small vector autoregressions, and leading indicator models. The method involves two steps: first, estimating the factors using principal components, and then using these factors to forecast the variable of interest. The estimated factors are referred to as diffusion indexes. The performance of the diffusion index forecasts is examined across different forecasting horizons (6, 12, and 24 months). The results indicate that the diffusion index forecasts, based on a small number of factors, perform well, with relative performance improving as the horizon increases. The improvement over the benchmark forecasts can be dramatic, in several cases producing simulated out-of-sample mean square forecast errors that are one-third less than those of the benchmark models. The paper also discusses the empirical factors and their characteristics, showing that a small number of factors can capture most of the forecasting improvement. The results suggest that a very small state vector may be necessary for forecasting macroeconomic time series. The paper concludes that the diffusion index method is a promising approach for macroeconomic forecasting.