A fake memory card reports a capacity it does not have, so it looks correct until the data you wrote is gone. Price is the tell you can use before buying: we compute a plausibility floor per capacity from live established-brand prices, and at 1TB anything under about $0.1178 per gigabyte should stop you. After it arrives, test the full capacity with h2testw or F3 before trusting it.
Floors recomputed from live prices · updated
The counterfeit that dominates this category is not a poor-quality card. It is a small card telling a large lie. Every memory card contains a controller chip that reports its capacity to whatever it is plugged into, and that report can be edited. A card holding 32 or 64 gigabytes of real flash is reprogrammed to announce 512GB or 1TB, and from that moment every operating system in the world believes it.
It mounts. It formats to the advertised size. The properties dialog shows the capacity you paid for. Files written inside the real capacity are stored and read back perfectly, which is why the first day, the first week and often the first month are uneventful. Past that boundary the card either silently discards what it is given or wraps around and overwrites data it already holds. There is no error, no warning and no indication until somebody opens an old file and finds it corrupt.
That delay is the whole design. A card that failed immediately would be returned; a card that fails in four months is usually blamed on the device, the user, or bad luck.
Everything else about a counterfeit can be copied. The packaging, the holographic sticker, the printed part number, the retail blister, the reported capacity - all of it is reproducible, and much of it is reproduced well enough that photographs will not separate them. What cannot be faked is the cost of the flash inside.
A genuine 1TB card contains roughly sixteen times the NAND of a genuine 64GB card, and that silicon has a floor price the manufacturer pays. A counterfeiter shipping 64GB while charging for 1TB has no such floor, and will price accordingly - deep beneath the genuine market, because undercutting is the entire sales pitch. So an implausibly low price is not a bargain that happens to be risky. It is the one part of the fraud that has to be visible.
| Capacity | Median $/GB | Below this, check | Observations |
|---|---|---|---|
| 64GB | $0.4959 | $0.2231 | 6 live listings |
| 128GB | $0.3124 | $0.1406 | 32 live listings |
| 256GB | $0.2656 | $0.1195 | 46 live listings |
| 512GB | $0.2265 | $0.1019 | 28 live listings |
| 1TB | $0.2617 | $0.1178 | 14 live listings |
Each floor is 45% of the median cost per gigabyte of the established-brand cards we currently track at that capacity. It moves when the market moves, which is the point - a threshold written into the page as a fixed dollar figure would be wrong within a season, and NAND pricing has moved a long way this year.
A floor is a screening threshold, not a verdict. Genuine clearance stock, a discontinued line and an aggressive legitimate sale can all appear beneath it. What crossing the floor means is that the listing has entered the price range where counterfeits operate, and that the remaining checks should happen before you order rather than after.
Who you are actually buying from. On marketplaces where several sellers share one product page, the reviews you are reading may describe a different seller's stock from the one about to ship. Check the seller name on the listing itself, not the brand at the top of the page. Where inventory is commingled, a genuine listing can still deliver counterfeit stock.
Capacities that do not exist. microSD ships in powers of two. A card advertised at 2TB in a category where the largest genuine cards are 1TB is not an early release.
Specification claims that contradict each other. A listing claiming a UHS-I card sustains 300MB/s is describing something the bus does not do. A listing claiming V90 on a budget card is claiming a certification that budget cards do not carry. Our speed-mark decoder covers what each mark actually guarantees.
Brand named only as compatibility. “For SanDisk”, “compatible with Samsung” and similar constructions appear when a seller wants a brand in the search index without claiming to be it. We reject these at seeding: the brand has to open the title, before any preposition.
Every check above is probabilistic. There is exactly one way to know: write data across the entire card and read it back. Two free tools do this and are the community standard.
h2testw - Windows. Fills the card with verifiable data and reads it back, reporting exactly how much of the claimed capacity is real.
F3 (Fight Flash Fraud) - Linux and macOS. The same test, open source, with its own maintained documentation.
We link these rather than reproducing their instructions. Both projects document their own use, those documents are maintained, and a copy on this page would quietly go stale. Run the test before you put anything irreplaceable on a new card - a full pass takes hours on a large card, and it is the only hours that will tell you the truth.
These pages list 128 cards. Getting to that number meant discarding listings that failed one of our gates, and the pattern of what failed is itself useful evidence about the category. From the seeding run on 17 August 2026: 129 candidate listings reached verification, 126 passed, 24 were rejected.
| 15 | No established brand at the front of the title The listing opened with an unrecognised or random-string brand, or named a real brand only as a compatibility claim - "for SanDisk" rather than SanDisk. |
| 3 | Not a microSD card at all A different card standard or a reader sold under memory-card search terms. |
| 3 | No live in-stock offer Listed but unbuyable at the moment we checked. |
| 2 | No plausible capacity in the title The stated capacity fell outside the range microSD actually ships in. |
| 1 | Accessory, not a card A case, adapter or holder rather than storage. |
The largest category by a distance is the brand gate. Fifteen of the twenty-four rejections were listings whose title did not open with a manufacturer we recognise - either an unfamiliar name, a string of characters, or a real brand invoked only as a compatibility claim. That is what the counterfeit-adjacent end of a memory-card search result looks like, and it is why the gate is applied at seeding rather than as a warning on a page.
One honest note on the price floor. In this particular run, zero listings were rejected by the plausibility floor after passing the other gates. That is not the floor failing - it is what happens when the brand and shape gates have already removed the listings that would have tripped it. The floor is the net under those checks, and it is published live above so you can apply it to any listing, including ones we never saw.
The controller inside the card is reprogrammed to report a capacity it does not have. A card holding 64GB of real flash can tell your computer it holds 1TB, so it mounts, formats and shows the right size in every properties dialog. Data written within the real capacity is stored normally. Data written beyond it is discarded, or overwrites what came before, and nothing warns you. The failure surfaces later, when files you saved months ago will not open.
Price is the most reliable signal available to you before purchase, because it is the one thing the seller cannot fake without losing money. A counterfeit ships a fraction of the flash it charges for, so it can undercut the genuine market by a margin no legitimate supply chain can match. The floors published on this page are computed from the established-brand listings we track, and a card well beneath the floor for its capacity is priced where a fake would be priced.
No, and we are careful to say so. A floor is a screening threshold, not a verdict. Genuine clearance stock, an older line being discontinued, or a genuinely aggressive sale can all sit beneath it. What the floor tells you is that this listing has crossed into the range where fakes live, and that you should apply the other checks before buying rather than after.
Test the full capacity before you trust it with anything. The community-standard tools write data across the entire card and read it back to confirm what was stored is what comes off - h2testw on Windows and F3 on Linux and macOS are the two everyone uses. We link them below rather than reproducing their instructions, because their own documentation is maintained and ours would not be.
It reduces the risk without removing it. Marketplaces that mix inventory from multiple sellers under one listing mean the product you receive is not necessarily the one whose reviews you read. Buying from the retailer directly rather than a third-party seller on the same page is the meaningful distinction, and it is worth checking which one you are actually buying from before you order.
Every listing must name an established manufacturer at the FRONT of its title, carry a capacity within the range microSD actually ships in, have a live in-stock offer verified against Amazon's catalogue API, and survive the computed price floor for its capacity. In the seeding run that built this catalogue, 24 of 129 candidate listings failed one of those gates. The breakdown is on this page.