In his book, "Home Machinist's Bedside Reader #2 (pg. 159)", Guy Lautard describes the "Osborne Maneuver" for accurately centering round stock in the milling machine using nothing more than an edge finder.
It works like this. Accurately measure the diameter of the stock. For description purposes, let's assume that the y axis is along the 12-6 o'clock line of the stock and the x axis is along the 3-9 o'clock line. Align the edge finder by eye to the 3 o'clock position and locate the edge of the workpiece. Now move by half the diameter towards the center of the stock along the x axis. Now, use the y axis controls to find the edge of the stock near the 12 o'clock position. Move half the diameter towards the center of the stock along the y axis.
Now do it again. Use the x axis controls to find the edge of the stock near the 3 o'clock position. Move half the diameter towards the center of the stock along the x axis. Use the y axis controls to find the edge of the stock near the 12 o'clock position. Move half the diameter towards the center of the stock along the y axis.
As you repeat this procedure again and again you will approach the center of the stock with ever increasing accuracy. (In mathematical terms, the procedure converges to the center of the stock.)
The question becomes, "How often do I have to do this?". The answer is, "Probably fewer times than you think!". I wrote OSBORNE.EXE to examine how fast the process converges. For example:
I got tired of rederiving the equations needed to make the calculations associated with this procedure so I wrote the included program. Given the description above, the program should be self-explanatory. If not, email me at mklotz@alum.mit.edu.
Here we have a 2 (we'll say inch but units don't matter) diameter workpiece and we initially aligned with an error of 0.1". That is to say, we initially aligned by eye to the x axis at the 3 o'clock position with an error of 0.1". If your eyes are that bad, you need better glasses! After the first iteration we're still 0.1" off the x axis (del1), but we're within 0.005" (del2) of being on the y axis. Our radial error (distance from the center of the workpiece) is the root-sum-squared of del1 and del2 or 0.100126". On the second iteration, del1 becomes the del2 of the previous iteration and that puts us within 0.0000126 on the x axis. The iterations continue in this fashion with del1 always becoming the del2 of the previous iteration.
As you can see, even with a hideous initial error we've converged to a nearly unmeasurable error after only three iterations. You can use the program to experiment with other combinations of workpiece diameter and initial error. Personally, I do it twice and don't worry about it.