The study of precursory phenomena shows increasing promise as a basis for earthquake prediction. Long-range forecasting can be expected to reduce by a large factor the uncertainty of estimates based on the historical record, and also to facilitate the development of short-range forecasting. The testing of prediction methods for reliability poses special problems. Earthquake forecasting will be much affected in practice by the social and economic implications of forecasts as such, and also by the relative implications of failures and false alarms, as well as successes.