Continued efforts to compare exponential smoothing with other alternatives to demand forecasting are summarized. Using stock-out risk at one extreme and oversupply at the other, the effects of variability in forecasting, even when accurate with respect to the mean, are highlighted. Using a normal model, exponential smoothing is identified as a major source of variability. Various forecast methods are compared using simulation relative to mean squared error when mean demand is allowed to vary according to specified patterns. In almost all circumstances, exponential smoothing consistently emerges as a first choice. The same alternatives are compared using real demand data and the results show exponential smoothing and maximum likelihood to be essentially equivalent.