How to Read a Label That Has Faded or Been Painted Over
Why this matters
An unreadable nameplate stops a job in a way almost nothing else does. You cannot order the right part, you cannot size the circuit, you cannot look up the print, and you cannot tell the customer what they own. Techs handle this badly in two opposite directions: they either give up in two minutes and start guessing from a similar unit, or they go straight at the plate with a solvent and a scraper and destroy the last readable characters on it. There is a recovery order that avoids both, and the single most useful rule in it is about when to stop climbing.
Before anything touches the plate
Nameplates live on and inside equipment, so the recovery method has to respect what the equipment is.
Electrical enclosures. If reaching or cleaning the plate means being at or inside an enclosure with exposed energized parts, open the disconnect and apply your lock and tag first, because 29 CFR 1910.333(b)(2) requires the circuit de-energized and locked or tagged before work on or near exposed energized parts, then prove it dead: meter on a known live source, meter on the conductors, meter back on the known source (NFPA 70E-2021, 120.5).
Fuel-burning equipment. Shut the appliance down and do not use an aerosol solvent or any flammable cleaner near a pilot, a burner, or an ignition source. If you smell gas at any point, that is a different job entirely: everyone leaves the building immediately, nobody touches a switch, nobody turns on a light, nobody uses a phone inside, and you call the gas utility from outside.
Pressure-retaining surfaces. Never grind, file, abrade, or punch on a pressure vessel shell or on a data plate attached to one. That plate is part of the vessel's required marking, and removing metal from it or from the shell is not a recovery method, it is damage to a pressure boundary.
The recovery ladder, least destructive first
Each rung risks more than the one above it. Work down, not in.
1. Change the light, not the label. Hold a flashlight almost flat against the surface so the beam rakes across it. Embossed, stamped, and even heavily faded printed characters throw shadows under raking light that are invisible under a straight-on beam. This costs nothing, risks nothing, and recovers more labels than every other rung combined.
2. Change your angle and magnify. Move your eye off-axis and use a loupe or your phone camera zoomed optically. Faded ink often survives as a texture difference rather than a colour difference, and texture reads at an angle.
3. Photograph and process. Shoot from several angles with the flash off-axis, then push contrast and drop to grayscale on the phone. A camera sensor separates tones your eye blends. Take the photographs even when you think you can read the plate, because they are the only version of it that survives the rest of this procedure.
4. Dry clean. Soft dry brush or a dry microfiber, no pressure, one direction. Removes dust and chalked surface oxidation without carrying pigment away with it.
5. Damp clean. Water with a mild detergent, applied to the cloth and never sprayed onto or into equipment. Test in a corner of the plate away from the data first. Wipe in one direction and check after every pass, because a printed plate can go from faded to blank in two wipes.
Stop here and run the parallel search before you go further. Rungs 6 through 8 can destroy the residual data you are trying to read, and the parallel search costs nothing and often finishes the job outright.
6. Take a rubbing. For anything embossed or stamped, lay paper over the plate and rub with the side of a pencil lead. Paint filling a stamped character actually helps here, because the relief is what you are reading, not the ink. This is the highest-yield rung for a painted-over stamped plate and it is still non-destructive.
7. Mechanical, edge first. A wooden or plastic scraper worked from an edge of the paint film, starting in a corner with no data on it, at a shallow angle. Never a metal blade on a printed plate. Stop the moment you see pigment coming off with the paint.
8. Solvent, last and tested. Applied to a cloth, never to the plate, tested first on a data-free corner, with the equipment shut down and no ignition source nearby. On an ink-on-substrate plate a solvent that lifts the paint usually lifts the printing with it, which means rung 8 can turn a partially readable plate into a permanently blank one. That asymmetry is the reason it sits at the bottom.
The parallel search: where the same data lives elsewhere
Run this the moment the plate resists rung 5. It is frequently faster than recovery and it produces better evidence.
| Source | What it gives you | How direct |
|---|---|---|
| Duplicate label inside the access panel or door | Often the electrical ratings and the wiring label | Direct, factory-applied |
| Data stamped or embossed in sheet metal or a casting | Model, serial, or a build code that survives paint | Direct |
| A major component's own label (motor, compressor, board, transformer, valve) | That component's ratings, and a build date | Direct for the component, indirect for the unit |
| Installer tag, permit sticker, inspection sticker | Model as recorded at install, install date | Copy, can be wrong |
| Panel schedule or circuit directory | Circuit and overcurrent device serving the unit | Copy, frequently stale |
| Prior service records, original invoice, permit record | Model and serial as recorded | Copy of a copy |
Rank what you find by how direct it is. A component's own plate is evidence about that component. A paper record is somebody's transcription, and transcriptions of a model number are wrong often enough that a single one never settles anything on its own.
What you never reconstruct
A model number can be inferred and then confirmed. A rating cannot be inferred at all.
Minimum circuit ampacity, maximum overcurrent protection, ampacity, pressure rating, temperature rating, gas input rate, and any other number that governs sizing or protection are recovered from a direct source or they are not recovered. Reading them off a similar unit down the row, or off a unit of the same family and a different size, is how a circuit ends up protected at the wrong value.
If a governing rating cannot be recovered, you have three honest options and no fourth: obtain it from the manufacturer using whatever identifier you did recover, design to the most conservative documented bound you can establish, or do not proceed with the work that depends on it.
A worked recovery: an outdoor unit painted twice
Condensing unit on a roof, adhesive nameplate faded by sun and overpainted along one edge during two building repaints. Disconnect opened, locked, tagged, live-dead-live proved on a known source before anything came near the panel.
Rungs 1 through 3. Raking light plus off-axis flash photography, contrast pushed, grayscale. Of the 11 characters in the model string, 7 come back legible and 4 remain in doubt. The 4 in doubt sit in the block that codes capacity and voltage, which is the common shape of a partial recovery: you get the family and lose the size. Knowing the family alone is not enough to order anything.
Rungs 4 and 5. Dry clean, then a damp wipe on a data-free corner. No additional characters. The plate is ink on a substrate and the ink is already at the edge of lifting, so the corner test is where this stops.
Stop and run the parallel search. The control panel cover carries a factory duplicate label with the minimum circuit ampacity and the maximum overcurrent protection printed on it, intact, because it lives under a cover and never saw sun or paint. That is a direct factory source for the two ratings that actually govern the electrical work, and it took about 0.1 hours to find.
For scale: the ladder to that point took roughly 0.3 hours. Both are absorbed effort hours on the same visit. The lesson is not that the ladder was wasted, it recovered 7 characters and the ratings would still have been needed. The lesson is that the parallel search should have been running from the start, not waiting at the bottom of a queue, because it was the step that ended the problem.
Closing the 4 unknown characters. The compressor's own label gives a build date and its own electrical data. Combined with the recovered 7 characters and the published nomenclature for that family, the candidate set narrows to a small handful of sizes. That is a narrowing, not an answer, and it gets written down as one.
Verifying a recovery before you rely on it
Two independent confirmations, or it stays labeled as inferred.
Independent means the two sources do not descend from each other. The nameplate remnant and a duplicate factory label are independent. A permit record and the customer's asset list usually are not, because one was typed from the other.
Then close it out properly. Fit a new durable label, and on it record each field with its source, not just its value: which fields were read off the original, which came from a duplicate factory label, and which were inferred and remain unconfirmed. Date it and initial it. A reconstructed nameplate presented as if it were read off the plate is worse than a blank one, because the next tech has no reason to doubt it and every reason to size from it.
References
- 29 CFR 1910.333(b)(2), de-energizing and locking or tagging circuits before work on or near exposed energized parts
- NFPA 70E-2021, 120.5, verifying an electrically safe work condition with the before-and-after instrument check on a known source
- Manufacturer documentation and published model nomenclature, which is the only sanctioned route from a partial model string to a confirmed one
- See related: Reading a Nameplate: What It Tells You; The Multiple Ratings on One Nameplate