Earthquake prediction is inherently statistical. Although some people continue to think of earthquake prediction as the specification of the time, place, and magnitude of a future earthquake, it has been clear for at least a decade that this is an unrealistic and unreasonable definition. the reality is that earthquake prediction starts from the long-term forecasts of place and magnitude, with very approximate time constraints, and progresses, at least in principle, to a gradual narrowing of the time window as data and understanding permit. Primitive long-term forecasts are clearly possible at this time on a few well-characterized fault systems. Tightly focuses monitoring experiments aimed at short-term prediction are already underway in Parkfield, California, and in the Tokai region in Japan; only time will tell how much progress will be possible.