As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
The accuracy figure on a thermal camera, typically plus or minus two percent of reading, describes the instrument’s own temperature computation over a stated ambient range. It does not describe the target. Emissivity, reflected background, distance and spot size each add error that the instrument cannot see. A camera with a published plus or minus two percent figure, such as the MiLESEEY TR20, is accurate only when the target fills the spot, has known emissivity, and sits in a stable ambient. Use the camera to find the anomaly; use a contact probe to measure it.
What the accuracy figure actually covers, and what it leaves out
A thermal camera does not measure temperature. It measures infrared radiance arriving at the lens, then converts that radiance to a temperature using three inputs: the emissivity you set, the reflected background temperature you set, and the distance you are from the target. The plus or minus figure on the box describes only the conversion arithmetic and the detector response, under laboratory conditions where those three inputs were correct.
Take the MiLESEEY TR20 listing, which publishes plus or minus two percent temperature measurement accuracy. That is a statement about the camera’s internal electronics. It says nothing about whether the wall you are pointing at is at the temperature the camera reports, because the wall’s emissivity is not two percent accurate. Most building materials sit between 0.85 and 0.95, but bare metal can be 0.05 to 0.2, and a shiny aluminium duct will read the room’s reflected temperature, not its own.
The error chain is multiplicative, not additive. A two percent instrument error on a reading of 60 degrees Celsius is about 1.2 degrees. A five percent emissivity error on the same reading shifts the result by roughly three degrees. A reflected background error of 10 degrees in a low emissivity target shifts it further. The instrument is the smallest term in the budget.
This is why a camera with a published accuracy figure and a camera without one can give the same answer on the same wall. The figure is a starting point, not a guarantee. For the physics behind the emissivity term, see what emissivity means in thermal imaging.
The distance problem: spot size, not zoom
Every thermal camera has an instantaneous field of view, the angular size of a single detector element. The spot size on the target is that angle multiplied by the distance. If the spot is larger than the thing you are measuring, the camera averages the thing with its surroundings, and the reading is a blend, not a measurement.
The listings here publish field of view but not the detector pitch or the IFOV directly. The FOXWELL RT280 states a 50 by 38 degree field of view and a 120 by 90 native detector, enhanced to 240 by 180 by its TISR processing. From those two numbers the native IFOV is roughly 50 divided by 120, about 0.42 degrees per pixel. At three metres that is a spot about 22 millimetres across. A 22 millimetre spot is fine for a wall, marginal for a cable lug, and useless for a surface mount component.
The enhanced resolution does not change this. TISR and similar super resolution schemes interpolate between native pixels; they make the image easier to read but they do not create new radiance samples. The spot size is set by the optics and the native detector, not by the display.
This is the single most common mistake in electrical inspection. A technician stands back to see the whole panel, sees a hot spot, and records a temperature. The recorded temperature is the average of the hot joint and the cooler busbar around it. The camera is not lying; it is reporting the average of what its pixels saw. Move closer, or use a camera with a narrower field of view, until the target fills several pixels.
Reflected temperature and the shiny target trap
Infrared cameras see reflected energy as well as emitted energy. A target with low emissivity reflects its surroundings strongly, and the camera cannot tell the difference between a photon that came from the target and a photon that bounced off it. The camera asks you to enter a reflected temperature, usually the ambient air temperature, and subtracts the reflected component. If you leave it at the default and the surroundings are not at the default, the reading drifts.
Low emissivity targets are the hard case. Bare copper busbar, galvanised steel, polished stainless, and aluminium foil faced insulation all reflect. The standard field fix is to apply a patch of known high emissivity tape to the target, let it equalise, and measure the tape. The tape reaches the target’s temperature by conduction and radiates like a black body. This is not a trick; it is the recognised method, and it is why electrical inspectors carry a roll of matt tape.
High emissivity targets are the easy case. Painted walls, plaster, wood, concrete, most plastics, and human skin all sit near 0.95 and are forgiving. This is why a phone attachment such as the FLIR One for Android USB-C works well for finding air leaks and moisture in a house, and poorly for checking a bare electrical joint without preparation. The physics does not change with the price of the camera.
For a fuller treatment of the emissivity setting and its effect on readings, see the emissivity guide.
How the listed cameras compare on the numbers they publish
The table below collects only the figures the supplied listings state. Where a listing is silent, the cell says so. Do not read a missing figure as a zero or a default; read it as unknown.
| Model | Native detector | Enhanced or display resolution | Field of view | Frame rate | Stated accuracy | Temperature range |
|---|---|---|---|---|---|---|
| MiLESEEY TR20 | 320 x 240 | Not stated beyond native | 50 degrees | 25 Hz | Plus or minus 2 percent | -4 to 1022 Fahrenheit |
| FOXWELL RT280 | 120 x 90 | 240 x 180 via TISR | 50 x 38 degrees | 25 Hz | Not stated in the listing bullets | Not stated in the listing bullets |
| FLIR One Android USB-C | 80 x 60 | 240 x 180 via Vivid IR | Not stated in the listing bullets | Not stated in the listing bullets | Not stated in the listing bullets | Not stated in the listing bullets |
| FLIR One Pro Android USB-C | 160 x 120 | 480 x 360 via Vivid IR | Not stated in the listing bullets | Not stated in the listing bullets | Not stated in the listing bullets | Not stated in the listing bullets |
Two things stand out. First, only one listing in this set publishes an accuracy figure at all, and that figure is the instrument term, not the system term. Second, the two FLIR phone attachments publish no thermal accuracy, no field of view, and no frame rate in their bullets; they publish resolution and software features. That does not make them inaccurate, but it means you cannot compare them on accuracy from the listing alone.
The native detector column is the one that governs spot size. The FLIR One Pro has twice the linear native resolution of the standard FLIR One, which means half the spot size at the same distance, which means it can measure smaller components without averaging. That is a real difference and it is stated. The enhanced resolution figures are display and processing claims, not additional radiance samples.
Where thermal accuracy stops being the right question
A thermal camera is a screening tool. It finds the place where temperature is different from its neighbours. It is very good at that, and it is fast, and it does not touch the target. It is not a substitute for a contact measurement when the number matters.
If you need to know the temperature of a bearing to decide whether it is failing, a thermal camera tells you which bearing is hotter. A contact probe or a thermocouple tells you how hot. The camera’s advantage is coverage; the probe’s advantage is a measurement that does not depend on emissivity, distance, or reflected background. The two belong in the same kit. The same logic applies when comparing a thermal camera with an infrared thermometer: the gun gives one spot reading with the same emissivity problem, the camera gives a map.
For confined space entry, gas detection, or any decision where a person’s safety depends on a number, a thermal camera is not the instrument. It does not detect gas, it does not measure oxygen, and it does not replace a calibrated meter. It can show you a cold spot that suggests a leak, but the confirmation comes from a different instrument.
And for measuring the temperature of something behind a wall, a thermal camera shows the surface. The surface is a boundary condition, not the interior. A borescope and a thermal camera answer different questions, and neither replaces the other.
The maintenance nobody mentions: the ambient reference
Every thermal camera has an internal reference, usually a shutter or a small black body the detector views periodically. The camera uses it to correct for detector drift as the electronics warm up. When the shutter flags, the image freezes for a fraction of a second; that is the camera recalibrating, and it is normal.
What is not normal is a camera that has been left in a hot vehicle or a cold van and is asked to measure immediately. The detector and the internal reference are not in equilibrium, and the first readings after power on can be off by more than the published accuracy. The fix is simple: let the camera sit in the ambient where it will be used for long enough to stabilise, usually fifteen to thirty minutes, and let it run its shutter cycles before you trust a number.
The other maintenance item is the window. A fingerprint on the germanium lens is not just a smudge; it is an emitting and reflecting surface in the optical path. Clean it with the method the manufacturer specifies, and check the image after cleaning. A dirty lens scatters energy and softens the very gradient you are trying to read.
Neither of these points is in the listing bullets for any of the cameras above. They are general to the technology, and they matter more than the difference between one accuracy figure and another.
The honest answer to the accuracy question
A thermal camera’s accuracy figure tells you how well the instrument converts radiance to temperature when the scene is described correctly. It does not tell you how well you described the scene. Set the emissivity, set the reflected temperature, get close enough that the target fills the spot, let the camera stabilise, and treat the reading as a screening value confirmed by contact where the number carries consequences. If you need a measured temperature rather than a located anomaly, the camera is the wrong instrument and a probe is the right one. For the wider selection picture, the thermal imaging camera roundup covers the field.
Frequently Asked Questions
Is plus or minus two percent good enough for electrical inspection?
It is good enough to find the anomaly. It is not good enough to certify a joint temperature without controlling emissivity and spot size. The two percent describes the camera; the scene adds more error than that. Use the camera to rank joints by temperature difference, then confirm the critical one with a contact probe or by applying high emissivity tape and re-measuring.
Why does my camera read a different temperature from my infrared gun?
Almost always because the two instruments are set to different emissivity values, or because they are averaging different spot sizes. The gun's spot is usually larger at the same distance, so it blends more of the surroundings. Set both to the same emissivity, measure the same patch from the same distance, and the readings should converge. If they do not, one of them needs calibration.
Does super resolution improve accuracy?
It improves the image, not the measurement. Vivid IR on the FLIR One and TISR on the FOXWELL RT280 upscale the displayed image from the native detector. The radiance samples are still the native ones, so the spot size is unchanged. A 120 by 90 detector enhanced to 240 by 180 still has the spot size of a 120 by 90 detector.
What is NETD and does it affect accuracy?
NETD is noise equivalent temperature difference, the smallest temperature difference the detector can resolve against its own noise. It affects how small a gradient you can see, not how accurately a single reading is converted. A low NETD camera shows finer thermal detail; it does not correct for emissivity. The TR20 listing states NETD under 40 millikelvin, which is a sensitivity figure, not an accuracy figure. See the NETD explainer for the distinction.
Can I calibrate a thermal camera myself?
Not in the sense of adjusting the accuracy. You can check it against a known black body source or a calibrated contact probe on a high emissivity target, and you can send it for calibration if it drifts. The emissivity and reflected temperature settings are user inputs, not calibration, and setting them wrongly is the most common cause of a reading that looks wrong.
When should I not buy a thermal camera at all?
When you need a measured temperature rather than a located anomaly, when the target is behind a surface, when the target is a gas, or when the decision is a safety one that requires a certified instrument. A thermal camera is a screening tool. If the job is to confirm a temperature to a standard, a contact probe or a calibrated meter is the right purchase, and the camera is an expensive way to avoid admitting that.