AbstractNatural disasters are caused by natural events, which can be grouped as (i) extreme unstable events; (ii) large ‘variability’ events associated with extremes of the inherent variability and chaotic behaviour of geophysical systems; and (iii) global change events. These primary events are usually followed by equally harmful secondary events, such as mud slides or tsunamis following volcanic eruptions. Whether these events lead to disasters depends to a large extent on the preparedness and resistance of the afflicted community – i.e. its vulnerability. Forecasts for these different categories of event are of two types: ‘risk’ forecasts based on data of previous events, and real‐time forecasts. For group (i) events, these forecasts only begin following some initial detection, whereas for groups (ii) and (iii), forecasts are made before any indication of the event. Many (but not all) types of secondary event may also be forecast with increasing accuracy using local geophysical data and computational models of relevant processes. In this paper the organisation of forecasts and warnings is described, as is how it involves geophysical and disaster or emergency centres at the national, regional and international level. Satisfactory arrangements for issuing, receiving and acting on forecasts and warnings between countries (including the use of broadcasting media) require carefully negotiated procedures at the intergovernmental level. It is pointed out that international procedures developed for meteorology need to be extended to include other types of geophysical event and disaster. The final section of this paper reviews the need for defining targets for improvement in forecasting during the current International Decade for Natural Disaster Reduction (IDNDR), and that this should be preceded by establishing the accuracy of current methods, i.e. the baseline. Clarification of our objectives and targets may well lead to greater financial support from governments and o