Long-lead seasonal forecasts of the 1997-98 El Nino and the 1999-2001 La Nina and their impacts (especially for cold season North America) were made with unprecedented skill, detail, and confidence. This success has led to enormous interest in seasonal prediction worldwide, including in developing countries, and in many instances unrealistic expectations about them. The fact remains that climate predictions have at best modest skill, and in many circumstances no or marginal skill, in the absence of a strong ENSO signal. Nevertheless, it has been argued forcefully by a number of groups that these sets of forecasts have tangible economic value for a class of decision-makers and users. The arguments are based on application of simple cost-loss models, which relate forecast value in a simple decision environment to various measures of forecast performance and skill. The credibility of the cost-loss arguments depends on whether real decision-makers and users behave in the manner assumed in the models. In existing applications of the models, the value of the forecasts is the mean expected over many consistent iterations of the forecast/decision process. In the case of a winter forecast, a particular user can only exercise this process once per year. Thus, for the cost-loss model value estimate to be relevant to the user, he/she must exercise the process over many years in a disciplined, consistent manner. It is difficult to imagine a user that would do this over a period of decades, or whose decision-making environment would remain static for that long. We present estimates of the uncertainty in forecast value for different forecast skills and realistic iterations of the forecasts/decisions, as well as with different simple assumptions about the psychology of the user (for example, the user abandons use of the forecast if the forecast was wrong two winters in a row). The computed uncertainties cast considerable doubt on the utility of marginally skillful forecast sets for individual users and provide a sense of what skill levels are necessary to increase likelihoods that relatively short sequences of forecasts/decisions will be of value. These skill levels turn out to be relatively high, comparable to the performance of the strong ENSO event "forecasts of opportunity." It is critical that potential users understand their cost/loss situation, the expected skill of forecasts on which decisions will be based, and what they must do to ensure a high likelihood of realizing value from use of the forecasts. The institutional production of seasonal forecast sets with marginal to modest skill is justified if, in aggregate, a substantial number of decision-makers for the forecast region uses each forecast.