Minimum Sample Weight: The Number Behind Every Result
Share
Every number on a certificate of analysis begins with a weighing. Before the column, before the titrant, before the calculation, somebody put a quantity of material on a balance and wrote down what it said. If that quantity was below the balance's minimum sample weight, the uncertainty of that single step can be larger than the difference the specification is trying to detect.
The subject is unglamorous and it explains a class of disagreements that otherwise look mysterious: two laboratories, same method, same lot, results a few tenths of a per cent apart, no instrument fault, no calculation error.
Readability is not accuracy
A balance that displays five decimal places reads to 0.01 mg. It does not measure to 0.01 mg. The display increment is a property of the electronics; the measurement uncertainty is a property of the whole mechanism, the environment and the operator.
The relevant performance figure is repeatability: weigh the same object ten times and look at the spread. That spread is typically several display increments, and it is the number from which the minimum sample weight is derived. A laboratory that quotes the display increment as its uncertainty has confused resolution with precision, which is the weighing equivalent of confusing a ruler's markings with its straightness.
Where 0.1 per cent comes from
The convention is that weighing error should not exceed 0.1 per cent of the reading. It is a convention rather than a law, but it is a sensible one, because it keeps the weighing contribution an order of magnitude below the tolerances that specifications usually work with.
The calculation is short. Take the repeatability of the balance, expressed as a standard deviation, multiply by a coverage factor, and divide by the target relative uncertainty. With a repeatability of 0.05 mg at the low end of the range and a target of 0.1 per cent, the minimum weight lands around 50 mg for a single weighing. Doubling the coverage factor doubles the minimum.
The practical consequence: weighing 20 mg on that balance puts the weighing error at roughly 0.25 per cent, which is two and a half times the intended contribution. Nothing fails visibly. The result simply carries more uncertainty than the report implies.
Why it matters more for assay than for impurities
An impurity result is a ratio of two small numbers and a percentage of the sample weight. A 0.25 per cent error in the sample weight moves a 0.10 per cent impurity to 0.10025 per cent. Nobody will ever notice, because the reporting precision is 0.01 per cent and the specification limit is a long way away.
An assay is different. The result is a percentage of a large number sitting close to a limit. A 0.25 per cent weighing error applied to a 99.6 per cent result can produce 99.35 or 99.85 depending on direction, and the first of those crosses a 99.0 per cent limit while the second makes the batch look better than it is. This is one of the several mechanisms behind an assay above 100 per cent, alongside residual water and solvent.
Titrimetric assays make the effect larger, because the titration consumes a volume proportional to the weighed mass, and the endpoint determination has its own uncertainty. The comparison with HPLC is worth reading alongside this note, because the two methods propagate weighing error differently.
The environmental part nobody records
Balances drift with temperature, air currents and vibration. A balance on a bench beside a door will show a different repeatability in February than in July, and the difference is large enough to change the minimum weight.
Three practical controls cost nothing: a draft shield that is actually closed, a location away from doors and windows, and a note of the ambient temperature at the time of the study. The third is what allows a later investigation to ask whether the number still holds in a different season.
Electrostatic charge is the fourth. It affects powders more than liquids, and a bile-acid powder in a dry laboratory in winter is a good candidate. The symptom is a reading that drifts upward for several seconds after the sample lands, and the fix is an ionising device rather than a longer wait.
Weighing by difference does not rescue a small sample
Weighing by difference is often proposed as a way to handle small quantities: weigh the container, add material, weigh again. It removes some systematic error from the container's mass, which is real. It does not remove repeatability, because two readings now contribute instead of one, and their errors combine.
In most cases the minimum weight for a difference weighing is slightly higher than for a direct weighing. The method survives because the quantity is comfortably above the threshold, not because the technique lowers it.
What a laboratory should be able to show
| Item | Why it is asked for |
|---|---|
| Repeatability study with date and balance ID | Establishes the number rather than inheriting it |
| The calculated minimum weight and the coverage factor used | Makes the assumption visible |
| The smallest weight actually used in the method | Shows whether the method sits above or below the threshold |
| Ambient conditions at the time of the study | Allows the number to be revisited in another season |
| Any weighing below the threshold and its justification | A deviation that was assessed is different from one that was not noticed |
None of these is exotic. All of them are cheap. The reason they are rare is that the problem is invisible until a result is challenged, and by then the sample has been consumed.
Where this leaves a supply-chain partner
We do not present an assay figure to two decimal places without knowing the weighing behind it, because the second decimal is often smaller than the weighing uncertainty and therefore decoration. Where a customer's method requires a smaller sample than our balances support with confidence, we say so and discuss the preparative dilution rather than reporting a number with a hidden error bar.
That conversation is more useful than a tighter-looking certificate, and it is one of the few places where a supplier can demonstrate competence before the audit rather than during it. The rest of the analytical chain, from validation to system suitability, rests on the same principle: the number is only as good as its weakest step.
A worked example, and what it shows
Take a balance with a repeatability standard deviation of 0.03 mg at a 20 mg load, and a coverage factor of two. The standard uncertainty of a single weighing is then 0.06 mg. Expressed relative to a 20 mg sample, that is 0.30 per cent. Expressed relative to a 100 mg sample, it is 0.06 per cent.
Now apply each to an assay result of 99.40 per cent. The 20 mg weighing carries an uncertainty of about 0.30 per cent relative, which puts the true value somewhere between roughly 99.1 and 99.7 per cent. The 100 mg weighing carries 0.06 per cent, which keeps the interval inside 99.3 to 99.5. Both certificates would read 99.4. Only one of them supports a specification limit of 99.0 per cent without argument.
The numbers above are illustrative and describe arithmetic rather than any particular batch or laboratory. The point they make is general: the reported value has a width, and the width is set by the smallest weighing in the chain.
Standards and samples are not weighed the same way
A reference standard is weighed once, and its weight propagates through every calculation in the run. A sample is weighed once per determination. That asymmetry means the standard's weighing error behaves as a systematic offset for the whole batch, while the sample's error behaves as scatter. A systematic offset does not average out with replicates, which is why the standard's weight deserves the larger quantity and the more careful technique.
It is also why an assay can be precise and wrong at the same time. Six replicates that agree to 0.1 per cent are describing the sample's weighing and the injection, not the standard's. If the standard was weighed at half the minimum weight, every one of the six replicates is offset by the same amount, and the agreement between them is a false comfort.
The public references for this are not obscure. The USP general chapters describe the balance performance expectations, the EDQM publishes the corresponding pharmacopoeial text, and the ICH quality guidelines place the whole question inside method validation, where the accuracy claim depends on it. A laboratory that treats the minimum weight as a number invented by an auditor has read none of them.
The question worth asking is not whether a balance is calibrated. It is what the smallest weight in the method is, and what the minimum weight for that balance is. If the first number is smaller than the second, the certificate is reporting more than it measured.
Designing the dilution so the weighing survives
Most methods fix the sample weight long before anybody checks the minimum. A standard dilution scheme might call for 20 mg into 100 mL, chosen because the concentration suits the detector rather than because the balance suits the weight. The two choices are separable, and separating them is usually free.
The sample weight can be raised fivefold and the volume raised with it, keeping the final concentration unchanged. The volumetric work grows slightly; the weighing uncertainty falls by a factor of five. Where the final dilution needs to be lower for the detector, a two-step scheme achieves the same concentration with the sample weighed at a comfortable level. The cost is a pipette step, and the gain is an assay whose second decimal place means something.
The same logic applies on the standard side, with the additional point made above: the standard's error is systematic, so its weight deserves the larger margin. A scheme that weighs 15 mg of sample and 60 mg of standard has the priorities reversed, and the reversal shows up as a bias rather than as scatter, which is the hardest kind of error to find after the fact.
None of this requires new equipment. It requires that somebody once wrote down the smallest weight the balance supports, and that the dilution scheme was checked against it. Both take an hour, once.
Frequently asked questions
What is the minimum sample weight?
It is the smallest quantity that can be weighed on a given balance so that the weighing error stays within a defined fraction of the reading, conventionally 0.1 per cent. Below that weight the balance's own repeatability becomes a visible part of the result rather than a rounding detail.
Does a five-digit balance solve the problem?
Only if its repeatability supports the weight in question. Readability is the smallest display increment. Repeatability is the spread of repeated weighings of the same object, and it is usually several increments. The minimum weight is calculated from repeatability, not from the display.
How does weighing affect an impurity result?
Indirectly but decisively. The impurity result is a ratio of two weighings, sample and standard. An error of 0.2 per cent in the sample weight moves a 0.10 per cent impurity result to 0.1002 per cent, which is harmless. The same error moves a 99.5 per cent assay to a value that can cross a 99.0 per cent limit, which is not.
Can the minimum weight be reduced by weighing by difference?
Weighing by difference removes some systematic error but not repeatability, because two readings are involved instead of one and their errors combine. In most cases the minimum weight for a difference weighing is slightly higher than for a direct weighing, not lower.
What documentation should a laboratory keep?
The repeatability study that established the minimum weight, the date, the balance identification, and the environmental conditions. Without that record the number is a habit rather than a finding, and habits do not survive an inspection.
Is this relevant for a 25 kg drum?
Not for the drum, but for the 20 mg taken from it. The weighing that matters is the one inside the method, not the one on the loading dock, and the two are separated by six orders of magnitude.