The short-term earthquake probability (STEP) forecasting model applies the Omori–Utsu aftershock-decay relation and the Gutenberg–Richter frequency-magnitude relation to clusters of earthquakes. It is mainly intended to forecast aftershock activity and depends on a time-invariant background model to forecast most of the major earthquakes. On the other hand, the long-range earthquake forecasting model EEPAS (every earthquake a precursor according to scale) exploits the precursory scale increase phenomenon and associated predictive scaling relations to forecast the major earthquakes months, years, or decades in advance, depending on magnitude. Both models are shown to be more informative than time-invariant models of seismicity. By forming a mixture of the two, we aim to create an even more informative short-term forecasting model. Using the Advanced National Seismic System catalog of California over the period 1984–2004, the optimal mixture model for forecasting earthquakes with M≥5.0 is a convex linear combination consisting of 0.42 of the EEPAS forecast and 0.58 of the STEP forecast. This mixture gives an average probability gain of more than 2 compared to each of the individual models. Several different mixture models will be submitted to the CSEP Testing Center at the Southern California Earthquake Center to ascertain whether or not this result is borne out by real-time tests of the models against future earthquakes.