Obtaining high-quality dispersion curves is a critical step in the development of reliable shear wave velocity profiles from surface wave measurements. Because of its limited equipment and space requirements, the refraction microtremor (ReMi) method has become a popular approach for determining surface wave dispersion curves, and is increasingly being used for estimating low-frequency (long wavelength) surface wave velocities that are beyond the range of most active sources. The recent development of a low-frequency field vibrator as part of the Network for Earthquake Engineering Simulation (NEES) program has made it possible to actively generate surface wave energy down to frequencies of less than 1 Hz. This paper presents a comparative study of the ReMi method and the active-source frequency-wavenumber (f-k) method (using the NEES vibrator) for developing low-frequency dispersion curves. Linear arrays of 1-Hz seismometers were deployed at eight deep soil sites in the Mississippi Embayment. Using both ambient and active energy, surface wave dispersion curves were determined to wavelengths of 600 m. The dispersion data from the two methods were in good agreement (within about ±5%) to wavelengths of 100 to 150 m (3 to 4 Hz) at most sites. However, at longer wavelengths the dispersion estimates from the ReMi approach deviated significantly from the active source measurements. The validity of the f-k dispersion curve was supported by dispersion data obtained using the SASW method, as well as ambient vibration measurements performed using a circular array at four of the sites. Analyses of ambient vibrations recorded using circular arrays show that the poor performance of the ReMi method at long wavelengths is not because of a lack of ambient surface wave energy at low frequencies, but can be attributed to invalid assumptions about the nature of the ambient wavefield.