Abstract A model for a strike-slip fault is proposed in which an instability (earthquake) and/or stable fault slip is produced by strain softening of the fault zone. The form of the fault zone constitutive law and the stiffness and dimension of the surrounding elastic plates determine the fault slip and plate strain as a function of time and the potential for instability. An instability is predictable from a developing plot of fault slip or plate strain if a single parameter, the maximum strain softening slope of the fault zone constitutive law is known. The amount of precursory fault creep and plate strain and hence the reliability and resolution of the prediction depends on the form of the constitutive law. If the rate of strain softening is rapid, precursors are virtually absent. For very slow rates of strain softening precursors are large but temporal resolution of the expected instability may be poor. The friction law on a fault segment may be inferred from observations of fault slip or surface strain as a function of time.