Previous geomagnetic field measurements in Colorado have been extended using three self-calibrating rubidium magnetometers accurate to about 0.01&ggr;. The magnetometers were deployed either in an L-shaped array or in a collinear array and provided simultaneous measurements of the scalar field at sites separated by 12--16 km. Previous noise reduction work using two-station magnetic field difference data from Colorado removed variations from the east-west scalar difference which were correlated with variations in the field component D. The earlier analysis work has been extended to remove also noise correlated with the field components H and Z. Noise due to susceptibility and magnetization differences between sites or induced currents in conductivity anomalies is reduced by this technique. However, the main purpose of this paper is to demonstrate a new noise reduction technique that removes noise due to large-scale gradients in geomagnetic variations. In effect, the method removes variations in the magnetic field difference between two sites which are correlated with variations in the difference between two other sites in a collinear array. This approach, which does not require vector data, reduced gradiometer noise to 0.03 &ggr; rms for 37 days of hourly averages.