What Is A Thermistor: Two Numbers And A Cheap Bead
A thermistor is a resistor whose value tracks heat. Two numbers describe it: R25 and beta.
The word is a squeeze of thermal and resistor. The part is a small bead, disc or chip of metal oxide, and it costs less than almost anything else on the board.
That cheapness is the trap. Two beads can look identical, sit in the same tape, and behave nothing alike. One is 10 kΩ in a warm room, the other 100 kΩ, and neither has a marking you can read.
This page covers the two families and the two numbers that define a part. It also covers the checks that stop a swap from ruining a reading.
The Short Answer
A thermistor is a semiconductor resistor whose resistance changes strongly with temperature. Change the temperature a little, and the resistance moves a lot.
There are two families. In an NTC part the resistance falls as the part warms. In a PTC part it rises.
NTC parts are the sensors. They read temperature, and they also limit inrush current when placed in series. PTC parts act as resettable fuses and as self-regulating heaters.
Most of this page is about the sensor side, because that is where the buying mistakes happen.
Two Families Under One Word
The two families are made from different materials and behave differently at the extremes.
NTC parts are sintered oxides of the iron group: chromium, manganese, cobalt, iron and nickel. Heat frees more charge carriers, so the resistance drops.
PTC parts are usually barium, strontium or lead titanates. Below the Curie point the material has a high dielectric constant and a low resistance. At the Curie point that constant collapses, barriers form at the grain boundaries, and the resistance climbs sharply.
Push a PTC part hot enough and it turns back into an NTC part. That reversal is real, and it is why a PTC is specified around its switch point rather than across the whole range.
There is also a silicon PTC, sometimes called a silistor. It is far more linear than the ceramic type and drifts less. It costs more, and it gives up the sharp switching behaviour.
| Property | NTC | PTC |
|---|---|---|
| Resistance as it warms | Falls | Rises, sharply near the Curie point |
| Body material | Sintered metal oxides | Doped barium or strontium titanate |
| Typical job | Temperature sensing, inrush limiting | Resettable fuse, self-regulating heater |
| The number on the label | R25 and beta | Switch temperature and hold current |
If the part is a PTC used as protection, the two numbers on its label are different ones. Ask for the hold current and the switch temperature, not for a resistance at 25 °C.
R25 Is Three Facts In One Number
Every NTC datasheet leads with something like “10 kΩ”. That figure carries three conditions, and dropping any one of them makes it meaningless.
The first is the temperature: 25 °C. Makers call it the standard reference point, and it is where the marked resistance is defined.
The second is zero power. The value is measured with very little current. Halve that current and the resistance may move by no more than 0.1 percent, or a tenth of the tolerance if that is smaller.
The third is tolerance. On disc and chip parts it commonly runs from ±1 percent to ±20 percent, and the width you accept is a price decision.
That is why “10 k” alone is not a part number. A bead at ±1 percent and the same bead at ±20 percent share a size, a colour and a price bracket. Only one of them holds a reading to a fraction of a degree.
Beta Is Not a Constant
The second number describes the shape of the curve. Beta, written B and measured in kelvin, says how fast the resistance falls as the part warms.
It is calculated from two points. The B equation relates resistance to temperature: 1/T = 1/T₀ + (1/B)·ln(R/R₀). Here T₀ is 298.15 K, and R₀ is the value there.
Read that again, because it explains a common argument. The result depends on the two points you picked.
One maker’s selection guide states its reference points plainly: 298.15 K and 348.15 K, which are 25 °C and 75 °C. Another supplier may quote 25 °C and 85 °C.
So B25/75 and B25/85 are different numbers for the same part. Comparing two datasheets without checking the pair of points is comparing nothing.
There is a second reason to be careful. Beta is a two-point fit, and the maker says outright that it is not the most accurate way to describe a curve.
The better fit uses three points instead of two. The Steinhart–Hart equation takes three coefficients, a, b and c, and holds the error under 0.02 °C across a 200 °C span.
| Model | Inputs | Where it is good enough |
|---|---|---|
| Beta, two points | R at two points | A narrow band around the points you fitted |
| Steinhart–Hart, three terms | R at three points | A wide span, error under 0.02 °C over 200 °C |
| Lookup table from the maker | A measured curve | Firmware, where the maths must be cheap |
For a working example, a 3 kΩ part at 25 °C takes a = 1.40×10⁻³, b = 2.37×10⁻⁴ and c = 9.90×10⁻⁸. Those three numbers sit in firmware, not on the order.
The order needs the two points behind the beta, because that is what makes two suppliers comparable.
The Same Bead, 10 k Or 100 k
Here is the failure that costs the most and looks the least dramatic. A production line runs out of sensors and a buyer substitutes a part that matches the drawing in every visible way.
It is the same bead diameter, the same lead length, the same epoxy colour. It is also 100 kΩ at 25 °C instead of 10 kΩ.
Nothing marks the difference on the part. The value lives in the tape label, the reel and the paperwork. That is the whole reason a loose bead in a bag is a purchasing risk.
The circuit then reads cold. In a divider that pairs the sensor with a fixed resistor, changing R25 by a factor of ten moves every reading on the curve.
Firmware fitted with the old coefficients makes it worse, because a correct reading is then converted with the wrong map.
Three checks catch this before it reaches the line. Measure the part at two known points and compare with the datasheet.
Ask for the resistance table the maker measured, not the one a catalogue page prints. And make the reel label part of the incoming record, so a swap cannot pass as a stock movement.
Two points are enough for the check. An ice bath sits at 0 °C and a still room sits near 25 °C. A part that misses both is not the part on the label.
Log the two readings against the datasheet, and the tolerance budget stops being an assumption.
How The Reading Becomes A Number
The sensor does not report a temperature. It reports a resistance, and the circuit turns that into a voltage.
The usual circuit is a divider: the thermistor on one leg, a fixed resistor on the other, and an analogue input reading the middle.
The fixed resistor sets where the divider is most sensitive. Pick it close to the value in the middle of your band, and the voltage moves fastest there.
This is the other half of the swap problem. Change R25 by a factor of ten and the divider sits at the wrong end of its range. The resolution collapses before the coefficients are even questioned.
A 100 k part in a circuit built for 10 k still produces a number. It is just a number with far fewer useful steps in the band you care about.
Self-Heating Makes Its Own Error
The bead is a resistor, so the current used to read it also warms it. The reading then sits above the true temperature, and the gap is yours to control.
The calculation needs one datasheet figure. The dissipation constant, given in milliwatts per degree, says how much power it takes to lift the body one degree.
So the error in degrees is the power you put in divided by that constant. A part that sheds 1 mW per °C, driven with 1 mW, reads about 1 °C high.
Two levers move that number. Use less current, or pulse the measurement and read between pulses. Both trade against noise, so there is a floor.
This is also why the drive circuit belongs in the specification. A sensor that is accurate on a bench and 2 °C high inside a sealed box was not a bad sensor.
For power parts the datasheet often states the maximum steady current at 65 °C ambient. Above that, the figure has to be re-rated rather than assumed.
Drift And Response Time
Two more figures decide whether the part still works in year five.
The thermal time constant is the time to cover 63.2 percent of a step in heat, measured with no power applied. A fat probe in a metal housing is slow. A bare bead in moving air is fast.
Drift is the slower worry. The same maker’s term list defines the zero-power value and the time constant precisely, and its selection guide puts a number on drift.
An epoxy-coated part can move about 0.2 °C per year. A hermetically sealed one moves about 0.02 °C per year.
Over ten years that is 2 °C against 0.2 °C. If the product claims ±0.5 °C, the cheap coating spends the whole budget on drift.
| Choice | Response | Drift per year | Fits |
|---|---|---|---|
| Bare or epoxy bead | Fast | About 0.2 °C | Consumer goods, short life |
| Glass encapsulated | Slower | About 0.02 °C | Appliances, instruments, wet areas |
| Potted probe in a housing | Slowest | Depends on the seal | Liquids, ducts, outdoor units |
Thermistor Or RTD
The other common sensor is the RTD, a resistance thermometer built from pure metal rather than ceramic.
The trade is simple. An RTD holds its accuracy over a much wider span. A thermistor is more precise inside a narrower one, roughly −90 °C to 130 °C.
Thermistor bodies usually work from −55 °C to +150 °C, and glass types reach +300 °C. NTC sensors are quoted from −50 °C to 250 °C, with the best accuracy between −50 °C and 150 °C.
So the question is not which is better. It is whether your product lives inside the narrow band where the cheap part wins.
| Property | Thermistor | RTD |
|---|---|---|
| Body | Ceramic or polymer | Pure metal |
| Range | −55 to +150 °C, glass to +300 °C | Much wider |
| Accuracy | Best inside a limited band, ±0.1 °C achievable | Holds over the full range |
| Curve | Strongly non-linear | Nearly linear |
One practical note on packaging. If the sensor sits in a wet area, the housing matters as much as the bead. How a sealed probe is rated tells you what the seal was tested against.
Six Lines for the Order
The order needs more than a resistance value. These six lines make it a part.
One. Family and R25 together with its tolerance, written as a percentage.
Two. Beta with both reference points, or the maker’s measured resistance table.
Three. The band that matters, and how close it must read.
Four. Packaging: epoxy or glass, lead length, and the housing if there is one.
Five. The drive current you will use, so the supplier can confirm the self-heating error.
Six. Whether the part must be curve matched or point matched. The first costs more and removes a per-unit calibration step.
Two background reads help before you send it. The line list a factory must fill in shows the shape of a brief that comes back comparable. The route a component takes through CE and RoHS covers the paperwork on the other side.
And if the sensor goes into a finished appliance, auditing the plant that assembles it is where you learn whether the swap rule is real.
Send me the datasheet or the drawing for the sensor you are buying. I will check whether R25 carries its tolerance, whether beta states its two reference points, and whether the drive current fits the dissipation constant. You get a short list of what to fix before the order.
Check The SensorQuestions Buyers Ask
What does 10 kΩ mean on a thermistor?
It is the resistance at 25 °C, measured with almost no power applied. The tolerance around that value is a separate figure, ranging from ±1 percent to ±20 percent.
Can I swap a 10 k for a 100 k part?
Only with a firmware change and a check of the divider. Ten times the resistance at 25 °C moves every point on the curve, and the old coefficients will read it wrong.
Why do two suppliers quote different beta values?
Because beta depends on the two points used. B25/75 and B25/85 are different numbers for the same part. Compare only after checking the reference points.
How accurate can a thermistor be?
Between 0.05 °C and 1.00 °C in general, and within 0.1 or 0.2 °C over 0 to 70 °C with tight parts. That accuracy depends on the drive circuit as much as the bead.
Does the sensor drift as it ages?
Yes. An epoxy-coated part can move about 0.2 °C per year, while a hermetically sealed one moves about 0.02 °C per year. Over a long product life that gap dominates the budget.
Why does my reading sit high?
Usually self-heating. The current used to measure the part warms it, and the rise in degrees equals the power applied divided by the dissipation constant in milliwatts per degree.
Questions & Comments
We read every one and reply within 24 hoursIf two of your sensors read a few degrees apart in the same bath, tell me which part numbers they are. The answer is usually in the label.