Abstract This article describes the results of a study of load forecasting at 49 of the 75 largest electric utilities in the United States. Historical forecasts and actual energy sales data were obtained from each participating utility along with descriptions of the technique or techniques used in the forecast preparation. Comparisons are made between techniques for forecasts with two, four, six, and eleven year horizons and vintages of 1972, 1976, 1978, 1980, and 1982. In addition, one multivariate and four univariate time series techniques were tested using annual sales data supplied by the utilities. The techniques tested were Univariate Adaptive Estimation Procedure (UNIAEP), Linear Regression, Holt's Exponential Smoothing, a combination technique which weights each of the previous approaches by one third, and a multiple regression approach where total electricity sales are forecast from state real per capita income, state population, and national real electricity price. The results show that the utility forecasts, especially end-use models, perform extremely well for the two year horizons but deteriorate over the longer term. Of the time series techniques tested, the combination technique and the Univariate Adaptive Estimation Procedure perform best over all horizons. Holt's Exponential Smoothing performs reasonably well in the short term while linear extrapolation performs fairly well over longer horizons. The results of the multivariate technique are disappointing in the short term but show some improvement for the longer horizons.