What scale do you need for making balms, and how do you check it is right?
A 0.01 g scale cannot weigh 0.01 g. Minimum weight from repeatability, worked for 1 g, 0.1 g, 0.01 g and 0.001 g scales, with the test weight class each one needs.
The scale you need is set by the smallest quantity you weigh, but not in the way the display suggests. A scale that reads to 0.01 g does not weigh 0.01 g, or 0.1 g, to any useful accuracy. What it can weigh honestly is a minimum weight, and that number comes from how much the reading wanders when you weigh the same thing ten times. Below: the arithmetic, the test, the scale each batch size needs, and the weight class that can check each scale.
Weigh any ingredient on a scale whose minimum weight is below it. For a 1 percent weighing error, minimum weight is 200 times the scale's measured standard deviation, and never less than 82 divisions. In practice a 0.01 g scale is good for about 2 g and up, a 0.1 g scale for about 20 g, a 1 g kitchen scale for about 200 g. Below your scale's minimum, buy a 0.001 g balance or make a premix. Check a 0.01 g scale with a class F2 weight; class M1 is too coarse for it.
Readability is not the smallest thing a scale can weigh
Three numbers describe a scale and the box only prints one of them. The readability, or scale interval d, is the step the display moves in: 1 g on a kitchen scale, 0.1 g on a coffee scale, 0.01 g on a jeweller's pocket scale. The repeatability is how far apart the readings land when you put the same object on the pan again and again. The minimum weight is the smallest net load at which repeatability is still a small enough fraction of the load to trust. The first is a property of the display; the other two belong to the scale on your bench.
The international recommendation for non-automatic weighing instruments, OIML R 76, gives every approved scale a minimum capacity, called Min, and explains why in one line: it is "specified to indicate that use of the instrument below this value is likely to give rise to considerable relative errors." The US equivalent, NIST Handbook 44, publishes the same threshold as a recommended minimum load "since the use of a device to weigh light loads is likely to result in relatively large errors." Both set it in scale divisions, and for the class of scale most makers use it is 20 divisions. On a 0.01 g scale that is 0.2 g.
Twenty divisions is a legal floor for trade, not an accuracy target. At 0.2 g on a 0.01 g display, one division is 5 percent of the reading. Anyone who needs the weight to be right needs a stricter rule.
Where minimum weight comes from: the 2000 times rule and its 1 percent version
United States Pharmacopeia general chapter 41, Balances, covers any substance that must be "accurately weighed". As the balance maker A&D summarises the chapter, repeatability is "assessed by measuring one test weight not fewer than 10 times", and it is satisfactory when two times the standard deviation, divided by "the desired smallest net weight", does not exceed 0.10 percent. Rearranged:
Minimum weight = 2000 x s, where s is the standard deviation of ten weighings.
If the standard deviation you measure is less than 0.41d, the chapter replaces it with 0.41d. A display that reads identically ten times in a row has not shown zero error; it has shown that its error is hidden inside the rounding. So under the chapter, the smallest possible minimum weight for any scale is 2000 x 0.41 x d, which is 820 divisions. On a 0.01 g scale, that is 8.2 g. Rice Lake's summary of the 2013 revision lists the same two changes: the expansion factor dropped from 3 to 2, and the 0.41d substitution was added.
A balm is not a tablet, and 0.10 percent is a pharmacy tolerance. Nothing in cosmetic law sets a weighing tolerance for a maker. This site works to 1 percent on each ingredient, the figure the rounding tables on batch to batch inconsistency and batch scaling are built around: 0.5 percent tocopherol weighed 1 percent light is 0.495 percent. Put 1 percent into the same equation and the constant drops by ten:
Minimum weight at 1 percent = 200 x s, and never less than 200 x 0.41d, which is 82 divisions.
Eighty-two divisions is the best case, for a scale whose readings never move. A scale whose readings wander by a division or two needs two to five times that, and the only way to know which you have is to measure it.
| Readability d | Legal Min (20d) | 1%, ideal scale (82d) | 1%, s = 1d (200d) | 0.10%, ideal scale (820d) |
|---|---|---|---|---|
| 1 g kitchen scale | 20 g | 82 g | 200 g | 820 g |
| 0.1 g coffee scale | 2 g | 8.2 g | 20 g | 82 g |
| 0.01 g pocket or bench scale | 0.2 g | 0.82 g | 2 g | 8.2 g |
| 0.001 g milligram balance | 0.02 g | 0.082 g | 0.2 g | 0.82 g |
Across a row, the legal floor and the pharmacy standard differ by a factor of 40 for the same scale. The working column for a maker is the middle one, 200 divisions, until you have measured your own scale and earned a smaller number. For comparison, A&D lists typical USP minimum weights of 14 g for its 2200 g laboratory balances reading 0.01 g and 1.4 g for its 220 g balances reading 0.001 g, measured "in a favorable environment".
Minimum weight applies to the net sample, not the gross load. USP chapter 1251, as summarised by Rice Lake, says so directly: "The minimum weight applies to the sample weight, not the tare or gross weight." Weighing 0.3 g of tocopherol into a 150 g jug is a 0.3 g weighing, whatever the display shows before you tare.
Measure your own scale: the ten weighing test
This takes ten minutes. Do it where you actually weigh, because draughts, heat from a hob and vibration are part of the result.
- Warm up and level. Switch on and leave the scale for the warm-up time in its manual. Level it if it has a bubble.
- Pick one object. A single stable object of a few percent of capacity, as A&D recommends. Rice Lake notes it "does not need to meet any calibration requirements", because the test only asks whether readings agree. A 10 g object on a 200 g scale is about right, and a coin will do.
- Weigh it ten times. Zero the empty pan, place the object in the centre, wait for the stable indicator, write the reading down, lift it off.
- Calculate the standard deviation. Any spreadsheet does it with STDEV (the sample form, dividing by n minus 1, which is the form A&D uses for its published figures).
- Apply the floor and multiply. If s is below 0.41d, use 0.41d. Multiply by 200 for a 1 percent minimum weight. Write the result on a label on the scale.
A worked example with illustrative readings from a 0.01 g scale. A 20 g weight read 20.00, 20.01, 19.99, 20.01, 20.00, 20.02, 19.99, 20.00, 20.01 and 20.00 g. The mean is 20.003 g. The sample standard deviation is 0.0095 g, almost exactly one division. That is above the 0.41d floor of 0.0041 g, so it stands. Minimum weight at 1 percent is 200 x 0.0095 = 1.9 g. Held to the pharmacopoeia's 0.10 percent, the same scale would need 19 g.
So this scale weighs 2 g of beeswax to 1 percent, and 0.25 g of tocopherol to about 8 percent, although no single reading strayed more than two hundredths from 20.00 g.
Run the test twice: once with the scale where it lives, once on the most solid surface in the house, away from the hob and any open window. If the second standard deviation is half the first, the scale is fine and the bench is the problem. The case for a dedicated weighing spot with nothing else on it is set out on workshop setup.
Which scale each batch actually needs
The table below works the common addition levels at four batch sizes and names the coarsest scale that can weigh each one to 1 percent, assuming a scale whose standard deviation is one division (the 200d column above). If you have measured your scale and it is better than that, the boundaries move in your favour.
| Addition level | 50 g batch | 100 g batch | 250 g batch | 1 kg batch |
|---|---|---|---|---|
| 20% (wax in a firm stick) | 10 g, 0.01 g | 20 g, 0.1 g | 50 g, 0.1 g | 200 g, 1 g |
| 5% (castor, a butter) | 2.5 g, 0.01 g | 5 g, 0.01 g | 12.5 g, 0.01 g | 50 g, 0.1 g |
| 1% (essential oil, flavour) | 0.5 g, 0.001 g | 1 g, 0.001 g | 2.5 g, 0.01 g | 10 g, 0.01 g |
| 0.5% (tocopherol) | 0.25 g, 0.001 g | 0.5 g, 0.001 g | 1.25 g, 0.001 g | 5 g, 0.01 g |
| 0.1% (pigment shade, an active) | 0.05 g, premix | 0.1 g, premix | 0.25 g, 0.001 g | 1 g, 0.001 g |
Three things fall out of the table. First, the 0.01 g pocket scale is a main-phase scale for small batches, not a minor-phase scale. It handles the wax and oil in a 50 g batch, not the essential oil. Second, a 0.001 g balance is the honest tool for anything under 1 percent of a batch under a kilogram, and the only alternative is the premix arithmetic below. Third, a 1 g kitchen scale is only a 1 percent tool from 200 g up, so it can weigh the oil phase of a kilogram batch and nothing in a 100 g one.
The rounding rules on the equipment page (a division no larger than a twentieth of the quantity) and the batch scaling page (readability 1 percent of the quantity) cover what the display can show. Minimum weight adds what it cannot: how far the reading moves between two identical weighings. Where they disagree, the larger number wins.
Below your minimum weight: premixes, and what they cost in accuracy
When an ingredient is lighter than your scale's minimum weight, dilute it until the portion you weigh is heavier. Pages here give the method for rosemary extract, synthetic antioxidants and ceramides, usually as a 1 in 10 dilution. The minimum weight tells you whether 1 in 10 is enough.
Work it for the measured scale above, with a minimum weight of 1.9 g, and a 100 g batch that needs 0.05 g of an active. A 1 in 10 premix makes the weighed portion 0.5 g, still below the minimum. Both the active and the portion need to be 1.9 g or more. Weigh 2.00 g of active into 78.00 g of carrier oil, which is a 1 in 40 premix of 80 g. Then 2.00 g of premix delivers 0.05 g of active. Take the 1.95 g of carrier out of the main oil so the formula still totals 100.
The premix is not free. Independent errors add in quadrature, so two weighings each good to 1 percent deliver a dose good to about 1.4 percent. It also leaves 78 g of premix on the shelf, with a shelf life of its own and a line in the batch record. A milligram balance weighs the 0.05 g directly.
Pigment shading is the case where this bites hardest. The worked tinted balm on iron oxides and mineral pigments needs 0.063 g of black oxide in a 60 g batch. That is below the 1 percent minimum weight of every scale on this page, even a milligram balance at its 0.082 g best, and that page notes that a 20 mg error in the black is a visible shift in the finished stick.
What an approved scale promises, and what a pocket scale does not
Approved laboratory balances carry two intervals: d, the step on the display, and e, the verification scale interval the scale is certified to. OIML R 76 allows e up to ten times d, and its own examples include d = 0.01 g with e = 0.1 g. The tolerances and the legal Min are stated in e, so that balance is a 0.1 g instrument as far as anyone has checked, with a legal Min of 50e (R 76, class II, e of 0.1 g or more): 5 g. Plan with e until your ten weighing test proves the extra digit is real.
Tolerances double between factory and workshop. R 76 Table 6 allows a new class II scale 0.5e up to 5,000e, 1e up to 20,000e and 1.5e above, and in service "twice the maximum permissible errors on initial verification". NIST Handbook 44 Table 6 gives the same in-service figures: 1, 2 and 3 divisions.
| Load on the pan | Divisions | Allowed error | As % of load at top of band |
|---|---|---|---|
| 0.2 to 50 g | up to 5,000 | ±0.01 g | 0.02% |
| 50 to 200 g | 5,001 to 20,000 | ±0.02 g | 0.01% |
| Above 200 g | over 20,000 | ±0.03 g | under 0.015% |
One division at 0.2 g is 5 percent of the load. A scale can be perfectly legal and useless for your smallest ingredient, because the law counts grams and you care about the share of what you weighed. R 76 also requires the same errors with a third of capacity off centre, so where you put the jug is part of the specification. A scale with no class and no approval marking promises none of this; the accuracy on its box is the seller's claim, which is a reason to test it.
Test weights: which class can check which scale
USP chapter 41 sets the ratio for check weights: the test weight's maximum permissible error, or its calibration uncertainty, must be "not more than 1/3 of the applied test limit". OIML R 111 uses the same third internally, requiring each weight's expanded uncertainty to be at most one third of its class tolerance. To know whether a scale reads within one division, the weight must be good to a third of a division.
R 111 classes run E1, E2, F1, F2, M1, M2, M3 from finest to coarsest; F2 and M1 differ by a factor of about three. A 100 g weight is allowed ±1.6 mg in F2 and ±5 mg in M1. A 200 g weight is ±3.0 mg in F2 and ±10 mg in M1.
| Scale readability | Check weight | Weight error allowed (d/3) | Coarsest class that fits | Class that does not |
|---|---|---|---|---|
| 0.001 g | 50 g | 0.33 mg | F1 (±0.3 mg) | F2 (±1.0 mg) |
| 0.001 g | 100 g | 0.33 mg | E2 (±0.16 mg) | F1 (±0.5 mg) |
| 0.01 g | 100 g | 3.3 mg | F2 (±1.6 mg) | M1 (±5 mg) |
| 0.01 g | 200 g | 3.3 mg | F2 (±3.0 mg) | M1 (±10 mg) |
| 0.1 g | 500 g | 33 mg | M1 (±25 mg) | M2 (±80 mg) |
| 0.1 g | 1 kg | 33 mg | F2 (±16 mg) | M1 (±50 mg) |
| 1 g | 2 kg | 333 mg | M2 (±300 mg) | M3 (±1,000 mg) |
An M1 weight is right for a 0.1 g scale and wrong for a 0.01 g one. An M1 100 g weight is allowed to be half a division out on a 0.01 g display, so when the scale reads 100.01 g you cannot tell whether the scale or the weight is responsible. Buy F2 for a 0.01 g scale, and buy it at a size near the top of what you actually weigh, not near the top of the scale's capacity.
A weight sold with no class marking and no certificate is a different thing again. R 111 is explicit that a weight whose errors exceed its class table "cannot be declared as belonging to the corresponding OIML class", and an unclassed weight has never been held to a table at all. It is a perfect object for the ten weighing test, and no evidence that the scale reads true.
Handle check weights with tweezers or a clean cloth, never fingers, and keep them in their case. Skin oil and dust are mass, and an F2 weight checking a 0.01 g scale is working within a few milligrams.
A calibration routine that fits a small workshop
Calibration in the strict sense is a check, not an adjustment. Rice Lake's summary of the USP revision says it plainly: "a calibration is a verification not an adjustment." Check and record first; adjust only when the check fails. A scale adjusted every morning without a record has no history, and the history is what shows drift before it ruins a batch.
- Before every batch, one reading. Warm up, zero, place the classed check weight in the centre, record the reading in the batch record. Ten seconds. If it reads outside one division, stop and run the full check.
- Monthly, the full check. Repeat the ten weighing test and recalculate the minimum weight; place the check weight at the centre and at four points around the pan, following the R 76 eccentricity idea with a load near a third of capacity; and read the check weight at two loads you actually use, for example 20 g and 100 g on a 200 g scale.
- Adjust only on a failure. Use the scale's own calibration mode, with the weight the manual names. The mode adjusts the scale to whatever you put on the pan, so an unclassed weight here makes the scale exactly as wrong as the weight.
- After any move, drop or new location, run the full check before the next batch.
- Write it down. Date, scale, weight, readings, standard deviation, minimum weight, whether you adjusted. The format that makes this part of the batch is on batch records.
Temperature belongs in the routine too. R 76 requires an approved scale with no stated working range to hold its errors between -10 C and 40 C, so in a balm workshop the temperature problem is local: a hot jug on the pan. Weigh solids and liquid oils cold where you can, and when a molten addition has to be weighed, the minimum weight you measured in still, cool air no longer applies. Thermometers need the same discipline; the ice point and in-range checks are on thermometer readings disagree, and the targets on pour temperatures.
Selling by weight: when the scale becomes a legal instrument
Once a net weight goes on a label, weighing becomes a legal matter too. The EU Non-Automatic Weighing Instruments Directive, 2014/31/EU, lists the uses for which a scale must meet its essential requirements, and one of them is "determination of price on the basis of mass for the purposes of direct sales to the public and the making-up of prepackages". Instruments approved for those uses carry the CE mark and the supplementary metrology marking, a capital M with the last two digits of the year, in a rectangle. NIST Handbook 44 lists class III as typical for "all commercial weighing not otherwise specified", with class II for laboratory and precious metal work.
How that applies to a maker filling their own tins depends on national law where the product is sold. The packers' rules for the average quantity system, the ℮ mark and the tolerable negative error are on labelling cosmetics in the UK and labelling cosmetics in the EU, and the checkweighing routine for fills is on filling equipment. Minimum weight applies there too: a 4.5 g lip tube checkweighed on a 0.1 g scale is a weighing below 20 g, and the scale cannot see the spread you are trying to measure.
Where this stops applying
Minimum weight measures scatter. It says nothing about a scale that reads consistently 1 percent high, which only a classed check weight finds, which is why the routine has both tests.
The 1 percent target is this site's working figure, not a regulation. A cosmetic safety assessment may set a tighter limit for a restricted ingredient, and where the concentration of a substance has a legal maximum, weigh it to the standard the assessor sets, which is covered on safety assessment and the CPSR. For essential oils, the dermal limits and the arithmetic per oil are on the essential oil dilution calculator.
The worked standard deviation here illustrates the method; it is not a measured property of pocket scales in general. The A&D figures are manufacturer typicals under good conditions. A weight sold as "F2" without a certificate is a claim, like the box on a pocket scale. For the arithmetic that turns percentages into grams once the scale question is settled, use the batch calculator, and for why a formula that worked at 100 g fails at 5 kg for reasons that have nothing to do with weighing, see formula fails after scale up.
Frequently asked questions
What scale do I need for making lip balm?
For the wax, butter and oil in a 50 to 100 g batch, a scale reading to 0.01 g, which weighs about 2 g and up to 1 percent if its readings wander by one division. For essential oil, tocopherol or pigment in that batch, a 0.001 g balance or a premix, because those additions are under 1 g and below a 0.01 g scale's minimum weight.
Can a 0.01 g scale weigh 0.05 g accurately?
No. One division is 20 percent of 0.05 g, and a scale whose readings wander by one division has a 1 percent minimum weight near 2 g. Even a perfect 0.01 g scale cannot go below 0.82 g at 1 percent, because the display rounds. Weigh 0.05 g on a milligram balance, or make a premix so the portion you weigh is over 2 g.
How do I work out my scale's minimum weight?
Weigh one object ten times, recording each stable reading, and calculate the sample standard deviation. If it is below 0.41 times the scale division, use 0.41 times the division instead. Multiply by 200 for a 1 percent minimum weight, or by 2000 for the 0.10 percent the United States Pharmacopeia uses for balances.
What class of calibration weight do I need?
Good to a third of one scale division at the load you check. For a 0.01 g scale that means OIML class F2 at 100 or 200 g; an M1 weight is allowed 5 mg at 100 g, half a division, so it cannot tell a good scale from a bad one. For a 0.1 g scale an M1 weight up to 500 g is enough.
Is the weight that came with my scale good enough?
For checking repeatability, yes, because that test only compares readings with each other. For checking accuracy or adjusting the scale, no: an unclassed weight has no stated tolerance, and a scale adjusted against it is exactly as wrong as the weight. Use a weight with an OIML class or a calibration certificate for that.
How often should I calibrate a scale for cosmetics?
Check it with one reading of a classed weight before every batch, and run the full check monthly: ten weighing repeatability, readings at the centre and four edge positions, and two loads you actually use. Adjust only when a check fails, and record every check, so a slowly drifting scale shows up in the record before it shows up in a batch.
What does the e value on a scale label mean?
It is the verification scale interval, the step the scale is certified to, as opposed to d, the step the display moves in. OIML R 76 allows e to be up to ten times d. A balance showing d = 0.01 g and e = 0.1 g is approved as a 0.1 g instrument, so plan with 0.1 g until your own test proves the last digit.
Sources and further reading
- International Organization of Legal Metrology, OIML R 76-1: Non-automatic weighing instruments, Part 1: Metrological and technical requirements, Tests, 2006, Tables 3 and 6, clauses 3.4.2, 3.5.2, 3.6.2 and 3.9.2.
- International Organization of Legal Metrology, OIML R 111-1: Weights of classes E1, E2, F1, F2, M1, M1-2, M2, M2-3 and M3, 2004, Table 1 and clause 5.2.
- National Institute of Standards and Technology, Handbook 44: Specifications, Tolerances, and Other Technical Requirements for Weighing and Measuring Devices, 2024 edition, section 2.20 Scales, Tables 3, 6, 7a and 8.
- A&D Company, What is minimum weight and how accurate weighing can be ensured, summarising USP general chapters 41 and 1251, with typical minimum weights by model.
- Rice Lake Weighing Systems, USP modifications in chapter 41 and 1251, comparison of the requirements before and after the 2013 revision.
- European Parliament and Council, Directive 2014/31/EU on non-automatic weighing instruments, Articles 1, 4 and 16, EUR-Lex.
Reviewed and updated 17 October 2026. Spotted an error? Tell us and we will fix and log it.