Think of "accuracy" versus "precision" when shooting groups. it applies to scales as well.
Your scale may be accurate (good repetition/consistency) but NOT precise (does not show correct weight)
In this case, it is almost impossible to duplicate your loads properly.
Your scale may be precise (shows correct weight) but NOT accurate (poor repetition/consistency, have to re-zero)
In this case, the scale will drive you crazy.
A "quality" scale should should have an acceptable, workable compromise of both, accuracy AND precision.
Someone else mentioned this but I figured I'd elaborate.
Precise means that you can hit the same spot every time (for a scale, that would mean if I put a weight that has not changed down on the scale 100 times, it should read within it's rated repeatability every single time).
Accurate means that you are hitting the spot you aim, on average, correctly (for a scale, this would mean that if I put a weight that is known to be 20 grains on the scale, it should show 20 grains, within the accuracy rating of that scale, every time).
A scale that is not accurate, but that is precise, is not a big deal for a hand loader, so long as he doesn't try to load those same loads on another scale and expect them to be as perfect as they were before. In practice, unless your scale is out significantly in it's accuracy, this shouldn't really matter much even then.
A scale that is not precise, but is accurate (which I find to be a bit of a misnomer, only because if you are accurate, then you are hitting the correct reading regularly, which suggests precision too, though technically doesn't mean precision because you could still be far off in repeatability though your average is accurate) will drive you nuts. You could put the same 20 grain weight on every time, and if your scale really is off, you could see 20.5 grains one time, 19.6 another, etc, and if it centres on 20.0 it could still be considered accurate, though also useless).
However, you should not have a compromise of both. A good scale will have excellent accuracy AND precision. They should all have ratings (any of the decent ones - a cheap dope scale likely won't actually fall within those ratings, but if you spend more than 80-100 bucks on a digital scale, it should have ratings that it adheres to pretty well). If it says it is accurate to 0.01 grains and has a repeatability of 0.01 grains, that means that if you put a weight on it that actually weighs 20 grains, you will get 19.99, 20.00, or 20.01 grains EVERY SINGLE TIME, so long as you have accounted for all outside variables such as wind, temperature drift, vibration, calibration if the scale was moved, etc.
I've got a Sartorious Master series AC211 (older analytical balance) which actually pulls this off even though it has very high resolution. That is an indication of, what I would consider, a GOOD scale. If it can show the digits, it should be able to resolve down to those digits. No point in a cheaper scale that reads down to 0.02 grains but is only accurate to 0.1 grains - might as well put tape over those last digits as they mean nothing on a scale like that.
Ebay is where I'd suggest looking for a good scale that you get for a good deal. The one I got is a great scale, albeit a little bit older, and it has a built-in calibration weight system which places and removes the internal calibration weight with a servo motor. You press calibrate and wait a few seconds, and you can hear it placing the weight, waiting while it calibrates, then removing that weight. Pretty slick system. Testing it afterwards with a high accuracy 100mg and 200 grain weight shows that it is dead on throughout that entire range. don't have a heavier calibration weight to test with, though the internal one is 150 grams and it too is dead on. When you're weighing to 0.1mg though, you've got to make sure you've taken wind, vibration, and even the surface into account. When I walk up to my scale, I can see the reading vary as the floor gives just ever so slightly, which angles the scale differently and produces a different reading. Very cool to see it be that accurate though.