Abstract.Investigations of the distribution of regional seismicity and the results of numerical simulations of the seismic process show the increase of inhomogenity in spatiotemporal distribution of the seismicity prior to large earthquakes and formation of inhomogeneous clusters in a wide range of scales. Since that, the multifractal approach is appropriate to investigate the details of such dynamics.Here we analyze the dynamics of the seismicity distribution before a number of strong earthquakes occurred in two seismically active regions of the world: Japan and Southern California. In order to study the evolution of spatial inhomogeneity of the seismicity distribution, we consider variations of two multifractal characteristics: information entropy of multifractal measure generation process and the higher-order generalized fractal dimension of the continuum of the earthquake epicenters. Also we studied the dynamics of the level of spatio-temporal correlations in the seismicity distribution. It is found that two aforementioned multifractal characteristics tend to decrease and the level of spatio-temporal correlations tends to increase before the majority of considered strong earthquakes. Such a tendency can be considered as an earthquake precursory signature.Therefore, the results obtained show the possibility to use multifractal and correlation characteristics of the spatiotemporal distribution of regional seismicity for seismic hazard risk evaluation.