Pipetting Accuracy and Volumetric Error in Serial Dilutions

A serial dilution is a chain of multiplications. Each tube's concentration is not measured; it is inferred from the tube before it and from the ratio of two volumes the analyst believes were delivered. Nothing in it is weighed or read back unless someone chooses to. That structure has a consequence protocol sheets rarely state: an error repeated at every step is compounded rather than averaged.

This guide covers systematic and random volumetric error, how each propagates down a dilution series, what gravimetric and photometric verification establish, and the arithmetic by which a tip error of two percent becomes a potency error of ten.

Two errors that need separate measurement

Volumetric performance is two numbers, not one. Trueness is how far the mean delivered volume sits from the set volume; precision is how tightly repeated deliveries cluster around that mean. A pipette delivering 9.80, 9.79 and 9.81 µL at a 10 µL setting is precise and untrue; one delivering 9.6, 10.4 and 10.0 µL is imprecise and, on average, true. The first defect survives replication; the second is beaten down by it.

In a study of a 96-barrel automated head verified spectrophotometrically against a calibrated manual pipetting reference, the best of four tested configurations — a 30 µL tip with a post-aspirate air gap, targeting 4 µL — gave a trueness of −2.22% bias with precision of CV ≤ 1% ^[3]. A bias several times larger than the scatter is the signature of a systematic offset, and no number of replicates will remove it.

How often either limit is breached was measured in a comparative study of 108 participants — 26 laboratory school students, 10 PhD students, 10 clinical chemists and 62 laboratory technicians — assessed by gravimetric replicates at 10 µL and 100 µL plus a serial dilution task using 1 M copper sulfate. Ten percent of participants exceeded ISO-defined accuracy limits at 10 µL and 22% did so at 100 µL, while imprecision limits were surpassed by 36% of participants at 10 µL and 20% at 100 µL ^[2]. The two failure modes did not track together across volumes, which is the practical argument for measuring them separately; the authors concluded that even experienced personnel occasionally failed to meet ISO standards ^[2]. A mini-review of manual pipetting makes the broader point that variation in pipetting is a known unknown: it is generally accepted that variations exist, but there have been limited studies of their extent in practice ^[1].

A dilution factor is a ratio, and that partly saves you

Consider a ten-fold series built by transferring 100 µL into 900 µL of diluent. The dilution factor is not the transfer volume; it is (100 + 900) / 100 = 10. Because it is a ratio, a bias that scales every volume by the same fraction cancels exactly. If both the transfer and the diluent are delivered 2% low — 98 µL into 882 µL — the factor is (98 + 882) / 98 = 980 / 98 = 10, unchanged.

What does not cancel is a bias confined to one of the two volumes. Take the same 2% low transfer, 98 µL, into a diluent volume correct at 900 µL. The realized factor is 998 / 98 = 10.1837, which is 1.84% larger than intended — negligible once, against most assay noise. Applied at every step of a six-step series it multiplies: 1.018367^6 = 1.1154. The realized dilution factor after six steps is therefore 11.5% larger than intended, which leaves the sixth tube at 1 / 1.1154 = 0.896 of its nominal concentration — a 10.3% shortfall in potency.

As one report on inline dilution quality control puts it, serial dilution methods are particularly prone to error propagation because each dilution is dependent on the previous concentration ^[4].

Random error grows as a square root, bias grows as a line

The two error types propagate on different curves. Concentrations multiply down the series, so relative errors add in logarithmic space. If each step carries an independent random relative error with a coefficient of variation of 2%, the accumulated CV after n steps is 2% × √n; after six steps, 2% × √6 = 4.9%.

Over the same six steps, the 1.84% systematic offset accumulated to a 10.3% shortfall in concentration. A bias smaller than the per-step noise therefore dominates the total error by the bottom of the plate, because n outruns √n. Replicating a serial dilution averages the 4.9% and leaves the 10.3% untouched. It is also the argument for parallel dilutions — each point prepared independently from the stock — where a transfer error affects one point rather than every point below it. The propagation algebra here is standard uncertainty propagation and is stated without citation.

Where the error actually lives: small volumes and real liquids

Several error sources are roughly fixed in absolute terms — the film retained on the tip wall, plunger backlash, judgment of the stop position. A fixed absolute error contributes a relative error proportional to 1/V, so a defect invisible at 200 µL is severe at 2 µL. The gravimetric data bear this out: median CVs across participant groups were 0.3% to 0.8% at 100 µL but 1.4% to 5.9% at 10 µL ^[2].

Liquid identity matters as much as volume. In structured pipetting exercises using water, methanol, serum and whole blood, with delivery determined gravimetrically from mass and literature density values for each matrix, low imprecision and bias were seen with water, while significantly increased imprecision and bias appeared in serum, whole blood and methanol; the authors concluded that water-based proficiency assessment leads to a false sense of technical ability ^[6]. Pre-rinsing — aspirating and dispensing the matrix three times to coat the tip — improved bias, particularly for methanol and whole blood ^[6]. Reconstitution solvents and viscous stocks are not water, and a pipette verified only on water has been verified on the easiest possible sample.

Gravimetric verification: what the balance does and does not tell you

Gravimetry converts a mass reading to a volume through a density; it is the reference method because mass is traceable in a way a plunger position is not. A mini-review of manual pipetting reviews gravimetry and spectrophotometry as the two common readouts for accuracy and precision, and provides protocols for single- and multi-channel pipettes suitable for competency training and regular checks ^[1].

A worked check. Ten replicate deliveries at a 10 µL setting give a mean net mass of 9.94 mg. Using a water density near room temperature of 0.9982 g/mL, the mean delivered volume is 9.94 mg ÷ 0.9982 mg/µL = 9.958 µL. The bias is (9.958 − 10) / 10 = −0.42%, where the denominator is the nominal set volume, not the measured one. If the ten masses have a standard deviation of 0.07 mg, the CV is 0.07 / 9.94 = 0.70%, and there the denominator is the measured mean. A single "accuracy" figure mixing those two denominators is a common way to make a pipette look better than it is. The density value used above is a tabulated physical constant, stated without citation.

Three limits are worth naming. Gravimetry measures what left the tip and landed on the balance, not what arrived in the destination well. It is slow at the replicate counts a meaningful CV needs. And at microliter and sub-microliter volumes, evaporation during weighing biases the result low unless controlled. A review of liquid volume measurement surveys the methods available from picoliters to a few milliliters and evaluates them on addressable volume range, attainable precision and accuracy, and suitability for in-line use ^[8] — an evaluation that reports each method's addressable range separately rather than identifying one technique that spans the whole interval.

Volume also leaves after dispensing. Small-volume microplate luminescence assays in an open-well format are described as very susceptible to sample evaporation, and one technique for reducing it layers silicone oil on the solution surface while still allowing automated reagent injection ^[9]. A well that has lost solvent is more concentrated than the arithmetic says — the opposite direction to a short transfer.

Verifying the series, not just the pipette

A pipette that passes a gravimetric check has not been shown to build a correct dilution series, because the series also involves mixing, tip changes and carryover. Photometric methods read the series itself. In one procedure for automated liquid handlers, a six- or seven-step serial dilution of Orange G was prepared manually in quadruplicate, each instrument pipette dispensed a selected volume eight times into wells brought to a common 200 µL total, and each pipette's dispensed volume was read from a plot of volume against absorbance; percent inaccuracy and percent CV were compared against predefined criteria, with failed pipettes repaired or compensated by a calibration curve in the instrument software ^[7]. The verification of the 96-barrel head described earlier used an Orange G protocol checked every 10 days over 30 days, agreed closely with the manual reference (r² = 0.998), and showed a slight tendency toward under-dispensing relative to manual pipetting ^[3].

Dye methods carry their own trap. The inline quality control report notes that assumptions about parameters such as dye linearity can lead to an erroneous process, and that the sulforhodamine 101 dye chosen for dimethyl sulfoxide systems behaves in a nonlinear fashion over its detectable range; the author handled this with a direct calibration including every point of the dilution template, tested across five identical instruments ^[4]. A calibration curve itself built by serial dilution inherits the error being measured.

Acceptance criteria are where convention shows through. A multi-author effort to establish quality assurance criteria for serial dilution operations set out four criteria any such process must pass — accuracy, precision, fold dilution and outliers — and describes thresholds that need to be both low enough to be acceptable from a biological potency variability perspective and high enough to let the instruments pass quality assurance analysis on a regular basis; the thresholds came from trend analysis of quality assurance data, a survey of screening scientists and chemists, and published single-shot criteria, with a mathematical analysis of the effect of threshold values on estimated XC50 values ^[5]. That is a negotiated standard rather than a derived one.

Instrument records are not evidence of delivery. A case report describes two instances of inappropriate pipetting by a vendor-programmed liquid handler in which the worklist obtained from the instrument failed to reflect what had actually been pipetted, and neither instance was caught during the laboratory's validation studies ^[10].

The solvent can change the concentration without anyone pipetting

Dimethyl sulfoxide is hygroscopic and absorbs water from the atmosphere quickly enough to matter on the bench. Time-course measurements found well geometry, fluid volume, well position and atmospheric conditions were all factors in hydration rate, with high rates in lower-volume fills and higher-density plates; in one reported example, a 1536-well plate filled with 2 µL of 100% dimethyl sulfoxide and exposed for one hour at approximately 40% relative humidity will absorb over 6% water by volume ^[11].

Treat that composition as a concentration change. If the dissolved solute and the dimethyl sulfoxide volume are unchanged and volumes are additive — both idealizations — a final solution that is 6% water by volume has a total volume of V(DMSO) / 0.94, so 2.00 µL becomes 2.13 µL and the concentration falls to 0.94 of nominal, a 6% drop before a single dilution step. The same source notes water in these solutions can accelerate degradation and precipitation ^[11]. Volatile solvents pose the complementary problem at the tip, consistent with the increased bias observed for methanol ^[6].

How a tip error becomes a potency error

A concentration error displaces a dose-response curve horizontally. A uniform error shifts every point equally, moving the midpoint without changing the shape. A compounding error changes the spacing between points, and spacing is what the slope is fitted to.

Suppose an eight-point series is intended at three-fold steps, spanning 3^7 = 2,187-fold, or log10(2187) = 3.34 orders of magnitude, but each step actually dilutes 3.1-fold. The realized span is 3.1^7 = 2,751-fold, 25.8% wider. If the analyst plots at nominal concentrations, each step advances 0.477 log units on the plotted axis while the material advances 0.491 log units in reality. The fitted response then appears to complete its transition over a log range 0.491 / 0.477 = 1.030 times narrower than the truth, inflating the apparent Hill slope by about 3% — assuming the top concentration is correct and the error accumulates purely multiplicatively.

Downstream estimators differ in how much of this they tolerate. Work on estimating microbial counts from a single dilution plate reported relative accuracy well within ±0.1 log10 in computer-simulated data, and stated that the method maintains this accuracy in the presence of dilution errors of up to 10% for both the aliquot and the diluent volumes, over counts from 10^4 to 10^12 colony-forming units ^[13].

Calibration curves are less forgiving. An examination of the imprecision in estimating PCR efficiency from standard curves, using six instruments, 1 to 16 replicates per concentration and transferred volumes of 2 to 10 µL, found by a Monte Carlo approach that the uncertainty in the efficiency estimate may be as large as 42.5% (95% CI) if a standard curve with only one replicate is used in 16 different plates; the authors also recommend that using a larger volume when constructing serial dilution series reduces sampling error ^[12].

What is not established

There is no agreed reference value for acceptable operator error. The manual pipetting mini-review argues that a common evaluation protocol would enable cross-laboratory comparison and could potentially facilitate the establishment of such a reference value — something the community might arrive at rather than something that exists ^[1]. Published acceptance thresholds for dilution series were set by negotiation against instrument capability and stakeholder survey as much as by biological argument ^[5]. And the liquid handler failure was not caught by the laboratory's own validation studies ^[10], which is a limit on validation as evidence, not a reason to skip it.

A minimal routine

The following is bench convention, stated without citation except where marked.

• Verify gravimetrically at the volumes actually used, not only at the top of the pipette's range, and report bias and CV separately with their correct denominators.
• Verify with the liquid class in use, since water flatters performance ^[6], and pre-rinse the tip for volatile or protein-rich matrices ^[6].
• Avoid transfer volumes below roughly 10% of a pipette's nominal capacity, which is convention rather than a measured threshold, and keep the transfer as large as the design allows, since using a larger volume when constructing serial dilution series reduces sampling error ^[12].
• Mix each tube to completion before drawing the next transfer, and change tips between steps.
• Where the series feeds a potency estimate, read it back photometrically rather than trusting the arithmetic ^[7] — remembering that the reading calibration may itself be nonlinear ^[4].