A new method for measuring arrival angles of teleseismic Love and Rayleigh waves is developed. The new method utilizes estimates of surface wave dispersion to create a phase-matched filter to isolate the Love or Rayleigh wave in three-component recordings. The polarization of the filtered wave group is determined in the time domain by application of a variation of the complex polarization method of Vidale <1986>. Orientation, linearity, and ellipticity of particle motion are estimated in several frequency bands to determine the frequency-dependent polarization. The method employs an iterative scheme, by which a predicted Love wave, based on the estimated dispersion and polarization, is subtracted from the three-component data prior to the estimation of Rayleigh wave polarization, and vice versa. The method is applied to an extensive set of Global Seismographic Network data covering the years 1989--1998. Between 4244 and 15,075 measurements are collected for fundamental mode Love and Rayleigh waves at nine different periods (37 to 150 s). Measurement uncertainties are estimated using the statistics of observations for pairwise similar paths and are generally of the order of 15--50% of the total signal, depending on the period and the wave type. Large and azimuthally invariant angle anomalies are documented for several stations and are consistent with misorientation of the horizontal seismometers. Two schemes are employed to determine the misorientations: (1) an azimuthally weighted average at each station, and (2) a joint inversion for seismometer misorientation and globally heterogeneous phase velocities. The determined corrections are robust and correlate well with those reported in earlier studies. Azimuthally varying arrival angle anomalies are shown to agree qualitatively with predictions of wave refraction calculated for recent phase velocity maps, which explain up to 30% of the variance in the new measurements.