Earthquake instability models have possible application to earthquake forecasting because the models simulate both preseismic and coseismic changes of fault slip and ground deformation. In the forecast procedure proposed here, repeated measurements of preseismic fault slip and ground deformation constrain the values of model parameters. The early part of the model simulation corresponds to the available field data, and the subsequent part constitutes an estimate of future faulting and ground deformation. In particular, the time, location, and size of unstable faulting are estimates of the pending earthquake parameters. The forecast accuracy depends on the model realism and parameter resolution. The forecast procedure is applied to fault creep and trilateration data measured near Parkfield, California, where at least five magnitude 5.5 to 6 earthquakes have ocurred regularly since 1881, the last in 1966. The quasi-static model consists of a flat vertical plane embedded in an elastic half space. Spacially variable fault slip of strike-slip sense is driven by an increasing regional shear stress but is impeded by a relatively strong patch of brittle, strain-softening fault. The field data are consistent with these approximate values of patch parameters radius of 3 km, patch center 5 km deep and 8 km southeast of the 1966 epicenter, and maximum brittle strength of 26 bars. Fluctuations in the available field data prevent estimating the earthquake time with any more precision than use of the 21¿8 year recurrence interval. However, the model may later give a more precise estimate of the earthquake time if the fault slip rate near the inferred patch increases before the earthquake, as predicted by the model.