I'll make a couple of comments on Richard's good suggestion. First of all, since a binocular hopefully retains one, single alignment independent of viewing distance and interpupillary distance, the images seen simultaneously through each of the two exit pupils with each of the two eyes will never exactly overlap except for one single distance (this is for horizontal axis overlap). I'm not sure what the distance is at which most makers want the optical axes to intersect, but a local binocular technician here in Finland collimates binoculars to intersect at ca 300 meters. In any case, do not expect perfect horizontal coincidence, especially if you look from as far back as 2 feet. However, what you see in the center of each exit pupil should not be very widely spaced in reality. If you are looking at an object one kilometer away, one degree would be ca 17.5 meters, and one minute of arc about a foot, or 30 cm. 5-10 minutes of arc of horizontal misalignment is generally considered pretty acceptable, although personally I feel that under five, or under 1.5 meters at one kilometer is what we should strive for.
When it comes to vertical misalignment, as Marc said, the standard should be stricter, and although some amount of vertical misalignment can perhaps be tolerated (perhaps about 2-3 feet/kilometer max), and is usually considered okay by the manufacturer's standards, zero would be best, especially since it is attainable. At least if we have had the binoculars re-collimated, we should expect to see virtually zero vertical "step".
Incidentally, in Sky&Telescope's comparison of seven binoculars for "stargazing", the best of the binoculars straight out of the box measured 4 arcminutes vertical misalignment and 7 arcminutes horizontal misalignment. These were all porroprism binoculars, but the price range was up to over a thousand dollars. Thus, none of them were even near perfect. Two had zero vertical misalignment, but had 15 and 20 arcminutes of horizontal misalignment respectively.
Roof prism binoculars have generally tended to be better collimated and retain their collimation better, but severely miscollimated samples are dishearteningly common among them as well.
Hope this helps you some.
Kimmo