Like many folks have said: I measure extreme spread. (There are other ways, such as plotting coordinates for each impact and computing variances... in other words, a lot more work!)
I bought a pair of calipers for the job... it makes things ever so much easier.
1) Measure from the extreme outside edge of the most widely-spaced projectile holes. With the calipers it is much easier to try all possible ways, and to get more digits of accuracy.
2) Subtract one diameter of a round. The answer will give you center-to-center measurements, despite calibre.
As
JohnKSa just pointed out, it is a good idea to measure a single, isolated hole in your target. The apparent size may differ from the stated calibre. You should also repeat this measurement for each new type of target paper, 'cause things change depending on the paper.
Now... here's a major factor to consider:
How Many Rounds are you going to include in each group???
The more rounds you include in each group, the smaller the variance will be between groups. Unfortunately, the group size will increase if you include more rounds in each group.
I normally use 5 rounds per group. That seems to be a widely accepted number, and it is a
LOT better than 3 rounds per group.
Unfortunately, my math says that your group sizes will vary greatly, even under identical conditions. For a 5-round group, and a 95 % confidence interval (look it up if you need to), you can expect your groups to vary over a range of about
* 4 *! That's right... do the same exact thing twice, and you could end up with 2 different group sizes... one of 'em
* 4 * times bigger than the other one!!! (actually, the difference could be worse... that's just the 95% confidence limit.) I've seen this kind of thing happen, which is one reason I started doing the math.
The way to get more consistency in your numbers is to send lots of rounds down-range and do averaging. I've done some math on this, and I've concluded that it's the total number of rounds that matters. Lots of groups with a small number of rounds each seems to be almost exactly as accurate as a few groups with lots of rounds each.
Now... for a really good time, start discussing the merits of MIL vs MOA measurements.