The discussion below is based on tests done with a new-style MG-BTA with a rotating head in August 2006.
The whole issue of accuracy is pretty difficult to address. There is no one easy way to state it. We can talk about the resolution of the unit on the two different scales, assuming a 12-bit A-to-D converter, as in LabPro, and the numbers are 0.004 mT and 0.0002 mT. We can talk about an average stability of the sensor (how much the reading changes over a minute or so, when you do not move anything. Those numbers are typically (0.04 and 0.005 mT). Or we could talk about the standard deviation of the readings over a minute when you do not move anything (typically 0.009 and 0.0007 mT).
Note also that magnetic fields are often affected by changes in the room. If you move a metal chair, you change its magnetic field. There can be many magnetic fields caused by electrical devices that cause readings to change.
In particular you should not expect to obtain a good reading for the magnitude of the Earth’s magnetic field due to the stray fields found indoors or near metallic objects.
A good strategy in using the field sensor is to fix the sensor in position, remove any moveable sources of magnetic fields, and zero the sensor. This will zero out the influence of background fields. Now, do not move the sensor, but move the magnetic field source you are studying.