I bought new bench scales (plus, discount code for Australian buyers)

[jfiresto]

Here are some school and entry-level lab balances that suggest the accuracy you can expect from a strain gauge load cell, with careful design and Chinese manufacture. Both do significantly better than 20 counts, worst case!

EDIT: I have an EMB 200-2, which is not bad, but I find a little too frugal with its just two buttons. I wish it had one more to skip the button dance you must do to start a calibration.

[Halcyon]

Well as I said, this is why I bought a decent set of scales this time. My previous ones have been crap and probably terribly inaccurate.

[HighVoltage]

My understanding is that it's a limitation with the load cells and the graduations are just one aspect of the "accuracy". The way it has been explained to me is that a worst-case, accuracy is about 20 times the graduation. So if you have a scale capable of 0.1g graduations, the maximum error you can expect (on a calibrated scale) is +/- 2 grams. I'm sure some other people here can explain the science/technology behind it.

If you're able to "tweak" the settings, it might just be that the chassis/firmware is designed for a number of different models (as is the scales I bought) however the load cells are physically different. You might actually be throwing your scales more out of spec by overloading them and potentially damaging the load cells.

Based on my limited experience with Mettler Toledo scales, the error is much smaller.
The linearity is just amazing and with the calibration weights I have and the resolution the scales offer, I can not detect a deviation from perfect mx+b function.
The calibration for these MT scales is done at 0 weight and full scale weight.

But one needs to be careful, not to move the scale after calibration.
You can easily detect a difference in gravitational pull, if you move a calibrated scale one floor up in the house.

Also, it is important to think about air weight displacement by the part that is being measured.
Look for "Balance Corrections for Buoyancy" see attached pdf

[Halcyon]

My understanding is that it's a limitation with the load cells and the graduations are just one aspect of the "accuracy". The way it has been explained to me is that a worst-case, accuracy is about 20 times the graduation. So if you have a scale capable of 0.1g graduations, the maximum error you can expect (on a calibrated scale) is +/- 2 grams. I'm sure some other people here can explain the science/technology behind it.

If you're able to "tweak" the settings, it might just be that the chassis/firmware is designed for a number of different models (as is the scales I bought) however the load cells are physically different. You might actually be throwing your scales more out of spec by overloading them and potentially damaging the load cells.

Based on my limited experience with Mettler Toledo scales, the error is much smaller.
The linearity is just amazing and with the calibration weights I have and the resolution the scales offer, I can not detect a deviation from perfect mx+b function.
The calibration for these MT scales is done at 0 weight and full scale weight.

But one needs to be careful, not to move the scale after calibration.
You can easily detect a difference in gravitational pull, if you move a calibrated scale one floor up in the house.

Also, it is important to think about air weight displacement by the part that is being measured.
Look for "Balance Corrections for Buoyancy" see attached pdf

As they should be. Like anything, you get what you pay for. I don't need lab-quality accuracy in the kitchen, but if I'm measuring a few grams, I want to be confident that the reading is accurate.

That being said, I wonder how much my calibration is off (if at all). They were calibrated at about 30 metres ASL, I'm at 800+ metres ASL.

[HighVoltage]

That being said, I wonder how much my calibration is off (if at all). They were calibrated at about 30 metres ASL, I'm at 800+ metres ASL.

For a "normal" scale, the error of +800m will not be detected.
At +800m you get around 0.01% deviation from your calibrated value.
A 1.0 gram weight would theoretically show 0.999748 gram at +800m


But the Buoyancy effect is much larger than I had thought.
For instance:

Steel has a density of 7850 kg/m3
A 2 kg weight 2*1000/7850 =  0.254777 liter of volume
1 m3 air is about 1.3 kg
1 liter of air is therefore 0.0013  kg
0.25 liter of air is 0.254*1.3/1000   =  0.0003302  kg  =  0.3302 gram

Therefore a 2 kg weight is about 0.3302 gram heavier in vacuum       
This is a difference of 0.0001651 or 0.01651 %  or 165 ppm
And it surprised me that it is so much.