Barcode failures are unusual among seller problems in that nothing tells you they happened. There is no rejection email and no error banner. A unit whose label will not scan simply does not become sellable inventory on the day it arrives, and you find out weeks later when you reconcile a shipment and the numbers do not match. The cost is never the label — it is the sell-through you lost while nobody knew the stock existed.
So this is a walk down the chain, from the moment the barcode exists as pixels to the moment a handheld scanner points at it in a fulfilment centre. Each stage below has one characteristic way of going wrong and one check that takes under a minute. Do the checks on a single label before you print five hundred.
Print scaling is the single most common cause
Every desktop print dialogue has a scaling control, and on most of them the default is not 100%. Adobe Reader ships with page-scaling behaviour that shrinks anything it considers oversized; browsers offer a “fit to page” option and frequently apply it without being asked. The result is a label printed at 94, 96 or 97 per cent — visually indistinguishable, and mathematically a different barcode.
It matters because a barcode is not a picture. It is a set of width ratios, and a scanner decodes it by measuring the narrow bar and comparing everything else to it. Shrink the whole symbol by 6% and the narrow bar shrinks too, which is survivable on its own. What is not survivable is what happens next, at stage two, when a printer has to decide which whole dots those fractional widths land on.
The same dialogue is where the quiet zone gets lost. That blank margin either side of the bars is part of the symbol rather than decoration, and a “fit to page” crop eats it first. If you are generating your own labels rather than using a marketplace-supplied PDF, the FNSKU and barcode generator exports ten modules of quiet zone on each side for exactly this reason, and it is why its output looks wider than the bars alone.
Print one sheet and measure a label with a steel rule. An Avery 5160 label is 2.625 × 1 in. If yours measures 2.5 in, the dialogue scaled it and every barcode on that sheet is suspect. Fix the setting, not the artwork.
A thermal printer cannot draw half a dot
Most label printers in a seller's back room run at 203 dpi, which means one dot is 1⁄203 of an inch — about 0.005 in, or 0.125 mm. Every bar and every space has to be a whole number of those dots. There is no in-between.
This is why a fractional scale is worse than a small one. Ask for a narrow bar of three dots and you get 0.0148 in, cleanly. Ask for 94% of three dots and the printer has to round 2.82 to either 2 or 3, and it will not necessarily round the same way for every bar in the symbol, because the rounding depends on where each bar falls on the dot grid. The output is a barcode whose bars are subtly uneven — some a third wider than their neighbours — and a scanner reading width ratios sees noise. Barcode software vendors document this directly: to get accurate symbols on low-resolution printers, the narrow-bar width has to be set as an integer multiple of the printer's dot size, and asking for a width the printer cannot draw simply produces the nearest one it can.
Print darkness is the second dial on the same machine. Too dark and heat spreads sideways into the white spaces, closing up narrow bars; too light and the bars are grey rather than black, which costs you contrast the scanner needs. Both errors look acceptable at arm's length and both are visible under a phone camera's macro mode.
Photograph a printed label with your phone as close as it will focus, then zoom in. You are looking for narrow bars that are all the same width and white gaps that are genuinely white. Ragged edges or gaps that have gone grey mean the darkness setting is too high.
Choosing between two printing technologies that look identical
Thermal label printers come in two kinds and the sticker they produce is indistinguishable on day one. A direct thermal printer has no ribbon; it burns an image into heat-sensitive paper. A thermal transfer printer melts pigment from a ribbon onto the label. The consumable cost and the print quality are similar. The lifespan is not.
Direct thermal media is, by construction, a material that turns black when it gets warm — so it keeps doing that after you print. Industry guidance puts direct thermal in the months-not-years bracket and reserves it for labels with short lives, such as shipping labels and receipts, while thermal transfer output is the recommendation for anything that has to stay legible in storage for a year or more. Heat, sunlight through a warehouse skylight, friction against another box and contact with oils or plasticisers all accelerate it.
For an FBA unit label that matters more than it first appears, because a unit label is not a shipping label. It is applied at your desk, and it may need to be readable when the unit is picked eight months later from a shelf, or when it is returned, or when it moves between fulfilment centres. That is a thermal-transfer or laser-sheet job. If you already own a direct thermal printer, this is not a reason to replace it — it is a reason to know that your labels have a shelf life and to avoid storing labelled stock somewhere warm.
Whatever the technology, the stock itself has to be matte. A scanner works by reading light bounced back off the label, and a gloss finish under overhead lighting throws a specular highlight straight into the sensor. Black on white, non-reflective, removable adhesive — those are the requirements the barcode generator page sets out alongside the label size window, and glossy photo-label sheets are the classic mistake.
Two seconds each. Rub a freshly printed label hard with your thumb — direct thermal will often smear or darken from the friction heat, thermal transfer and laser will not. Then tilt it under a ceiling light: if you can see the light source reflected in the label, the stock is too glossy.
Where the sticker goes changes whether it can be read
A barcode scanner needs the whole symbol in one plane. Put a label across the corner of a box and the bars on either side of the fold are at different angles to the sensor, which is a decode failure on a perfectly printed label. The same applies to a sticker wrapped around a curved bottle, one placed over a seam or a flap that lifts, and one applied to shrink wrap that has wrinkled — a wrinkle running across the bars compresses their apparent width exactly like a bad print does.
Choose the flattest available panel, avoid the last quarter-inch of any edge, and if the packaging is a soft poly bag, get the label onto the flat face rather than the gusset. On a bottle or tumbler, the label needs to sit along the axis where the curve is gentlest, and if there is no such panel, the barcode may need to be smaller so that it spans less of the curve.
This is also the stage where the other rule lives: when a unit carries a marketplace-issued code, that label has to cover or obscure every other scannable barcode on the package — the manufacturer's UPC, an ISBN, a courier sticker left over from the inbound freight. Placing the new code tidily beside the old one is the tidy-looking version of the mistake. The reasoning, and the difference between the two labelling schemes it comes from, is on the generator page.
Label one real unit, not a flat sheet, and scan it with a free phone barcode app from about a foot away at a slight angle. Then compare the string the app returns against your intended code, character by character. A generator will happily render a typo into a flawless barcode of the wrong product.
The label has to survive the journey, not just the printer
Between your desk and a scanner there is a carton, a pallet, a lorry and a shelf. Labels get scuffed against neighbouring units, condensation gets into cartons that move between temperatures, and adhesive that was fine on cardboard peels off the polypropylene of a poly bag after a week. None of that shows up in a test print.
Two habits deal with most of it. First, apply labels to clean, dry, dust-free packaging — adhesive failure is nearly always a surface-contamination problem, and a fingerprint of hand cream is enough. Second, if your product ships in something abrasive or is likely to sit in storage, a clear protective laminate over the label is cheap insurance; it is a common recommendation for exactly this reason, though it needs to be genuinely clear and matte, since a glossy overlaminate reintroduces the reflection problem from stage three.
The last habit is record-keeping rather than physics. Keep one printed sample per SKU from every label run, dated. When a receiving discrepancy appears three weeks later, the difference between a five-minute diagnosis and an unanswerable question is whether you can look at what you actually printed.
Take a labelled unit, put it in a box with the rest of the shipment, shake it, take it out and scan it again. If a two-second shake defeats the label, a two-week journey certainly will.
What to do when a label is already out there and failing
If units have shipped and the shipment is not reconciling, the useful move is to work out which stage failed, because the remedies are different. A whole shipment failing points at stages one to three — something systematic in the file, the printer or the stock, and the fix applies to every unit you have ever printed that way. A scattered handful failing points at stage four or five, which is a handling problem rather than a printing one.
Either way, the sample you kept is the evidence. Scan it. If your own retained sample scans cleanly and returns the right string, the print chain was sound and the damage happened after the label left you; if it does not, you have just reproduced the fault at your desk, which is much better news than it sounds.
One thing this guide deliberately does not do is restate marketplace label policy. Label dimensions, permitted symbologies, adhesive type and what must appear on the sticker alongside the barcode are all set by the marketplace, they differ between marketplaces, and they change. Amazon's own labelling documentation sits behind a Seller Central sign-in, so confirm the current requirements for your marketplace there rather than from any third-party page, including this one. Everything above is about the physics between the file and the scanner, which does not change when a policy does.
The related concern, if you are setting up a new SKU rather than relabelling existing stock, is that a correctly labelled shelf of units and a suppressed listing produce the same commercial result. Where the detail cut from a shortened title is meant to go covers the listing side of the same launch, and the reason a printed label carries a resolution figure at all — that 300 in “300 DPI” is a property of the file, not the printer — is worth two minutes if a sheet ever comes out soft enough to misread.
Written 2026-07-28. Printing behaviour described here is a property of the hardware and does not carry a policy date; marketplace label requirements do, and are not restated above.
Sources: the dot-quantisation problem and the advice to set narrow-bar width as an integer multiple of the printer's dot size are documented by IDAutomation's barcode knowledge base — adjusting narrow bar width on low-resolution printers and barcodes not scanning from a 203 dpi printer. The direct-thermal versus thermal-transfer lifespan comparison follows label-industry guidance, including Brother Mobile Solutions and Continental Datalabel, which place direct thermal in the under-six-months bracket and recommend thermal transfer for anything held in storage for a year or more. Label size, symbology, adhesive and content requirements for FBA units are Amazon's and are set out in Seller Central; they are referenced here but not reproduced, because they vary by marketplace and change without notice.