Before the days of cameras with multiple focus points, it was common to focus on the object of interest, lock focus and then recompose. However from time to time you see comments in photography forums that this process can lead to focus errors.
So what's the truth? Does focusing and recomposing lead to focusing errors? The answer is yes...and no! Yes, in theory (and in practice) there will be a focusing error, but no in the sense that the error may be small enough that you'll never notice it.
To explain this I'll use the diagram below which shows how a rectilinear lens (and most photographic lenses that aren't fisheye lenses are rectilinear) forms an image on a sensor. The Lens (shown in gray) images a flat plane (shown in red) onto a flat sensor or film (shown in blue). As you can clearly see, the points L (left) and R (right) are further from the lens than the point C, which is in the center and on which the lens is focused.
If the distance at which the lens is focused on "C" is given by "d" as shown in the figure, then the distance to the points "R" and "L" will be given by x = d/cos(theta), where theta is the angle between "C" and "R" (or "L"). If points "R" and "L" are at the edges of the frame, theta is half the horizontal coverage angle of the lens.
The distance "x" will always be greater than the focus distance "d", so if the focus is fixed and the camera is turned so that, say, point "R" is in the center of the frame and point "C" is at the edge ("focus and recompose"), the lens will be focused at a distance "d", but the object in the center of the frame will be at a distance x = d/cos(theta).
Similarly, the original subject in the center "C", will still be at the original focus distance "d" from the lens, but the plane of focus at the edge of the frame will be at a distance of x (= d/cos(theta)) from the lens. So the original subject will now be slightly in front of the focus plane (a distance of [d/cos(theta) - d] in front).
So let's take an example. Let's take a 50mm lens on an APS-C Digital camera like the EOS 30D. The horizontal angle of view will be 25.4 degrees (See my article and calculator for Field of View), so theta will be half that, 12.7 degrees. Lets say that the lens is focused on point "C" at a distance of 10m. If we lock focus and recompose so that point "R" is now in the center of the frame and "C" is at the edge, the original point "C" will now be (10/cos(12.7) - 10)m in front of the focal plane. That works out to about 0.25m (25cm).
But will it matter? Lets say we're shooting wide open at f2. What's the depth of field at 10m. The answer is the depth of field will extend from 8.7m to 11.7m (see my Depth of field article and calculator). So even at f2 the amount of defocus will be very small compared to the depth of field and "focus and recompose" should result in a sharp image.
So when won't it be OK? Well, the greater the angle though which the camera is turned, the greater the focus difference, so the wider the angle of view of the lens, the greater the possible focus difference will be. The faster the lens and the closer the focus, the smaller the depth of field will be. So the worst case would be a close focused fast wideangle lens, shot wide open.
Let's look at a 20mm f2.8 lens on a full frame 35mm camera and focused at a distance of 1m. A 20mm lens has a horizontal angle of view of 84 degrees, so the largest angle you could turn the camera through to focus and recompose would be half that, or 42 degrees. You'd be focused at 1m, but an object at the edge of the frame which was originally in the center would be about 0.35m in front of the focal plane. Depth of field would be from 0.8m to 1.3m, so in this case focusing and recomposing would result in a significantly out of focus subject. But 20mm at f2.8 and focused at 1m is a pretty extreme situation.
If you run through all the numbers and make assumptions about typical subject distances, apertures and angles likely to be swung through when focusing and recomposing you find that the vast majority of the time it shouldn't give you a significant loss of sharpness. In fact most of the time it probably won't be detectable.
Here's an example. I took a Canon EF-S 10-22mm lens and focused on a test chart at a distance of about 8" from the front of the lens at 10mm and f3.5. Using an EOS 20D I first focused using the center AF point. Then I locked the focus by switching to manual and rotated the camera until the target was underneath the leftmost AF zone, which is about 1/2 way to the edge of the frame. That's a rotation of approximately 24 degrees. I then took a second shot. Finally I switched the lens back to AF, refocused and took the third shot. 100% crops from the resulting images are shown below:
The first shot (shown on the left), with the target and focus in the center of the frame, is clearly the sharpest of the three as would be expected. With the lens rotated (but not refocused) so the target was halfway to the edge (shown in the center) you can see there is some "distortion". This isn't actual distortion in the technical sense. It's the result of the rectilinear mapping of the lens and would be seen with any lens of the same focal length no matter how good (or expensive!) it was. Sharpness has dropped a little as a result of the aberrations present in most wideangle zooms. The final (rightmost) image shows what happens when the lens is refocused. As you can see there is a very slight increase in sharpness. These are 100% crops and so represent what you'd see in something like a 16x20 print (maybe bigger, depending on your monitor resolution), so the difference is very small.
In a second example I used an EOS 20D with a 50/1.8 lens focused at a distance you might typically use for a headshot, about 4ft (1.2m). The lens was used wide open at f1.8.
As you can see there's essentially no difference between sharpness at the center (left) and the sharpness with the focus locked at the center, then the camera turned though an angle to put the same target 50% of the way to the edge of the frame (right).
With longer lenses used further from the subject the effect would be even less. So I think it's pretty safe to say that focusing and recomposing, even though it theoretically can result in being slightly off focus, works pretty well in practice and it's not something you really need to worry about 99% of the time.
Note that this analysis assumes a flat field. In practice not many lenses have a flat field. Depending on which way the field curves, results may be better or worse than the flat field case would predict.