Ant,

Looks to me like Steve and Andy are hinting that a noise floor of 90 microvolts is equivalent to 70 milligauss (so no you can't measure down to 2 milligauss unless you characterize a particular serial number sensor and then place a heavy low pass filter on the output -

90 microvolts / 1.30 millivolts per gauss (nominal) = 70 mgauss

Consider the noise bandwidth for the device - 10 Hz to 10 kHz - (DWG A-12, 505). Is your application in that range of frequency? Can you modify the setup to present a different range of frequency excitation to the sensor? Will the sensor respond to the frequency you've chosen?

Consider the effect of temperature (DWG A-12, 573). What options exist to eliminate measured effects of temperature drift? Put the sensor in a temperature stabilized environment, compensate for the variation by adding a voltage bias which a function of temperature, adding an external nulling coil to force the sensor to measure zero gauss despite the applied field?

Consider supply voltage stability (DWG A-12, 506 through 508). How stable is your supply voltage? Is the supply stable with the expected temperature and load changes?

Finally, consider linearity (DWG A12, 509). This curve doesn't really show you a transfer characteristic of output voltage vs gauss. To get a sense of what's being stated, get a sensor and play with it.

Cheers,

Bruce