we can calculate orientation (from -90° to 90°) on one axes by

asin(val);

where val HAVE to be from -1 to 1, so

maxValue = 1024

half=maxValue/2; //because we have value from 0 to 1024, 0 is -G, 1024is +G and 512 is 0G

val = (rawRead-half)/half; //the /zero assure that we will have a result in range of -1, +1

(notice you have to use acos if this axis is parallel to gravity vector, so asin for x and y, acos for Z)

now we have to set our precision. with a precision of +-4g, 1g = 512/4 = 128LSB so 1LSB = 1/128 G

precision now is asin(0) - asin(1/128)

asin of 0 is 0 (how convenient!)

asin of (1/128) is 0.44°, so 0.44° is your precision

but we don't need 4g, because any value above 1G is just noise from acceleration, so we can use the lowest resolution witch is +-2G, 1G= 512/2=256LSB, so 1LSB=1/256

asin(1/256) = 0.22°

Okay, thanks for that. I guess I was more off the mark than I thought.

The number to look for is the sensitivity figure. This particular chip has the max sensitivity ratings of 256lsb/g, so each "bit" corresponds to 1/256g. or 90/256 ~= 0.4 degree, in theory.

In reality, you would be lucky to get to within 2-3 degrees.

That is unfortunate. I may have to look for a more accurate sensor in that case.