We present a new approach to polarization analysis of seismic noise recorded by three-component seismometers.It is based on statistical analysis of frequency-dependent particle motion properties determined from a large number of time windows via eigenanalysis of the 3-by-3,Hermitian,spectral covariance matrix.We applied the algorithm to continuous data recorded in 2009 by the seismic station SLM,located in central North America.A rich variety of noise sources was observed.At low frequencies (<0.05 Hz) we observed a tilt-related signal that showed some elliptical motion in the horizontal plane.In the microseism band of 0.05-0.25 Hz,we observed Rayleigh energy arriving from the northeast,but with three distinct peaks instead of the classic single and double frequency peaks.At intermediate frequencies of 0.5-2.0 Hz,the noise was dominated by non-fundamental-mode Rayleigh energy,most likely P and Lg waves.At the highest frequencies (>3 Hz),Rayleigh-type energy was again dominant in the form of Rg waves created by nearby cultural activities.Analysis of the time dependence of noise power shows that a frequency range of at least 0.02-1.0 Hz (much larger than the microseism band) is sensitive to annual,meteorologically induced sources of noise. We present a new approach to polarization analysis of seismic noise recorded by three-component seismometers. It is based on statistical analysis of frequency-dependent particle motion properties determined from a large number of time windows via eigenanalysis of the 3-by-3, Hermitian , spectral covariance matrix. We applied the algorithm to continuous data recorded in 2009 by the seismic station SLM, located in central North America. A rich variety of noise sources was observed. At low frequencies (<0.05 Hz) we observed a tilt-related signal that showed some elliptical motion in the horizontal plane. The microseism band of 0.05-0.25 Hz, we observed Rayleigh energy arriving from the northeast, but with three distinct peaks instead of the classic single and double frequency peaks. -2.0 Hz, the noise was dominated by non-fundamental-mode Rayleigh energy, most likely P and Lg waves. At the highest frequencies (> 3 Hz), Rayleigh-type energy was again dominant in the for m of Rg waves created by nearby cultural activities. Analysis of the time dependence of noise power shows that a frequency range of at least 0.02-1.0 Hz (much larger than the microseism band) is sensitive to annual, meteorologically induced sources of noise.