Abstract An analysis of existing earthquake precursor data leads to a conclusion that the precursors reported so far can be classified into three types, i.e., A1, A2 and B types. Most of type B precursors, observed in terms of anomalous tilts and strains or foreshocks, have no magnitude-dependent precursor time. Meanwhile the A2-type precursors observed by means of geodetic work, changes in seismic-wave velocities and the like seem to have a precursor time which is closely correlated to the magnitude of coming earthquakes. A precursor of this type may possibly be interpreted by the current theory of dilatancy. The A1-type precursors, observed mostly several hours prior to the main shock, may be caused by a creep-like failure before the main rupture of the Earth's crust. Probabilities for an anomalous signal of various geophysical elements to be related to a forthcoming earthquake are estimated on the basis of the existing data of precursors. A feasible strategy for predicting a large earthquake as suggested by the present study will be as follows: First of all, we monitor accumulation of crustal strain by means of geodetic work. The next task is to detect an A2 signal which would arise from a highly strained crust sooner or later. If the spatial extent of the A2-type precursor is known, it is possible to estimate roughly the magnitude as well as the occurrence time of the coming earthquake. Finally, detection of an A1-type precursor, if it should occur, would provide a short-range forecast having a time span of hours.