A novel acquisition and processing technique is described to derive surface-wave dispersion curves from seismic ambient noise. We have determined that the use of spatial gradients of the wavefield provides new opportunities for high-resolution near-surface characterization with minimal field effort. In contrast to conventional active source data analysis that provides spatially smoothed results from large and dense arrays of receivers, our method provides local phase velocity information from ambient noise using only three closely spaced geophones. A time-frequency domain polarization-based analysis scheme is implemented to (1) detect the useful part of the data satisfying fundamental gradiometry assumptions, (2) estimate the noise source azimuthal distribution, and finally, (3) derive the desired local dispersion curves. The robustness and effectiveness of the developed method is demonstrated using the synthetic and real data, with a comparison to results from active source measurements.