We investigate the potential of two Independent Component Analysis (ICA) methods, i.e., the temporal and spatio-temporal ICA, for separating geophysical signals in GRACE data. These methods are based on the assumption of the statistical independence of the signals and thus separate the GRACE-observed mass changes into maximal independent signals. These two ICA methods are compared to the conventional Principal Component Analysis (PCA) method. We test the three methods with respect to their ability to separate a periodic hydrological signal from a trend signal originating in the solid Earth or the cryosphere with simulated and CSR GRACE mass changes for the time period of January 2003 to December 2010. In addition, we investigate whether the methods are capable of separating hydrological annual and semi-annual mass variations. It is shown that both ICA methods are superior to PCA when non-Gaussian mass variations are analyzed. Furthermore, the spatio-temporal ICA resolves successfully the lack of full temporal and spatial independence of the geophysical signals observed by GRACE both in global and regional simulation scenarios. Although the temporal and spatio-temporal ICA are nearly equivalent, both superior to PCA in the global GRACE analysis, the spatio-temporal ICA proves to be more efficient in regional applications by recovering more reliable the postglacial rebound trend in North America and the bimodal total water storage variability in Africa.