Rockbursts, as they occur in South African deep level gold mines, have a lot in common with natural earthquakes. Both violently release seismic energy and cause damage to underground and surface structures. Rockbursts are the main cause of fatalities in deep gold mines and the problem may increase in severity with future mining at greater depths. Although the emphasis in Rock mechanics is on the prevention of rockbursts, prediction would provide a partial solution to the rockburst problem. A research project was initiated in 1981 on Western Deep Levels to investigate the possibility of detecting and identifying precursive microseismic activity. This project has succeeded in locating potential sources of seismic events with reasonable confidence. The timing of these events may be estimated with less confidence, and an important parameter, namely the potential size of the rockburst, still cannot be determined. An example of a large seismic event and its associated precursive activity are discussed. Practical prediction systems will, at least for some time, involve subjective decision making which will place a new responsibility on the mine manager as well as the rockburst specialist. The moral pressure on the manager as the final decision maker will probably be greater than the fear of leqal prosecution.