Wavenumber analyses in MR elastography (MRE), which use first-order finite difference operators, are known to be more stable against noise than second-order finite derivative methods. However, wavenumber analyses for the human brain normally suffer from abundant heterogeneities and solid-fluid interfaces. We here present multifrequency wavenumber analysis for MRE of the human brain in 2D and 3D based on adapted bandpass filters. We show that both approaches provide better repeatability in test-retest experiments compared to standard analyses. Moreover, wavenumber analyses yield stable values and rich detail in regions of lower signal-to-noise-ratio such as deep gray matter.