The critical point hypothesis for large earthquakes predicts two different precursory phenomena in space and time, an accelerating moment release and the growth of the spatial correlation length. The objective of this work is to investigate both methods with respect to their predictive power. A systematic statistical test based on appropriate random earthquake catalogs allows to quantify the correlations of a precursory pattern with the subsequent mainshock activity. The analysis of target earthquakes in California since 1960 with magnitudes M ≥ Mcut reveals that these correlations increase systematically with growing Mcut, and correlations at greater than 95% confidence are observed for Mcut ≥ 6.5 in the case of the spatial correlation length. In particular, the seismicity patterns are found to be significantly correlated with each of the largest earthquakes (M ≥ 7.0), individually. The acceleration of the moment release has a similar trend, but is less significant.