Thermocouple vs RTD vs Thermistor: How to Choose

A kiln controller, a cold-chain logger and a battery pack monitor all measure temperature, and they almost never use the same sensor. The choice between thermocouple vs RTD vs thermistor is usually explained with a three-row table: thermocouples for hot, RTDs for accurate, thermistors for cheap. The table isn’t wrong, but it hides the errors that actually decide a design: cold-junction compensation for thermocouples, lead resistance and self-heating for RTDs, and nonlinearity for thermistors. This article explains how each sensor works, puts numbers on those errors and gives a way to choose that starts from your range and your error budget.

Key Takeaways

• Range decides first. Type K thermocouples are specified from −270 to 1,370 °C; RTDs and thermistors cover far less.
• A type K thermocouple produces about 39.5 µV/°C near 0 °C. Its reading is only as good as the measurement of its cold junction.
• A Pt100 RTD changes about 0.39 Ω/°C. Ten meters of two-wire cable can add more than 4 °C of error; three- and four-wire connections remove most of it.
• An NTC thermistor can change by around 4 % per °C at room temperature (with a typical B value), hundreds of times the RTD’s change in ohms, but its curve is strongly nonlinear.
• Standard tolerance classes differ a lot: a class B Pt100 is ±0.8 °C at 100 °C; a class 2 type K thermocouple is ±3.75 °C at 500 °C.

“Which temperature sensor should I use?”

The question appears constantly in engineering forums, and the answers usually start with “it depends”. The things it depends on are concrete.

What range does the measurement need? Above about 600 °C, thermocouples are usually the only practical option among the three.

How accurate does the reading need to be, and over how long? A process that must hold ±0.5 °C for years has different needs from a fan controller.

How far is the sensor from the electronics? Cable length and routing matter differently for each sensor: microvolts on a long thermocouple run pick up interference, ohms of lead resistance add directly to a two-wire RTD.

How much signal processing is available? A thermistor read by a microcontroller with a lookup table is cheap; a thermocouple read to 0.1 °C needs a low-noise converter and a cold-junction sensor.

How each sensor works and where its errors come from

Thermocouples: two metals and a reference you have to measure

A thermocouple is two dissimilar metals joined at the measuring point. The temperature difference between that junction and the point where the wires connect to copper (the cold or reference junction) produces a small voltage: the Seebeck effect. Joseph Wu’s thermocouple guide for Texas Instruments gives the type K sensitivity as 39.48 µV/°C at 0 °C, and lists the specified ranges of the common types: type K from −270 to 1,370 °C, type J from −210 to 1,200 °C, type T from −200 to 400 °C and type S up to 1,768 °C.

Two consequences follow from the physics. First, the signal is tiny. Resolving 0.1 °C with type K means resolving about 4 µV, which is why the TI guide recommends low-noise delta-sigma converters. Second, the thermocouple measures a difference. The electronics must know the cold-junction temperature, usually with an RTD, thermistor or diode on the terminal block. Any error in that measurement goes straight into the result, roughly degree for degree.

The TI guide makes a further point that trips up many firmware implementations. The voltage-to-temperature relation is nonlinear, so the correct method is to convert the cold-junction temperature to its equivalent thermocouple voltage, add it to the measured voltage and convert the sum to temperature. Adding the two temperatures directly gives an error; the guide’s type K example shows 1.5 °C. The reference functions for that conversion are published by NIST in Monograph 175 and adopted in IEC 60584-1.

Thermocouples also have the loosest standard tolerances of the three. Under IEC 60584-2, as tabulated in the TI guide, type K class 1 is the larger of ±1.5 °C or ±0.4 % of the temperature, and class 2 is the larger of ±2.5 °C or ±0.75 %. At 500 °C, class 2 allows ±3.75 °C.

RTDs: platinum, a quadratic and the wiring around it

An RTD is a platinum resistor whose resistance rises with temperature, described by the Callendar–Van Dusen equation with coefficients standardized in IEC 60751. Callendar showed the near-quadratic behavior of platinum in 1887, and it is why platinum remains the reference for accurate industrial temperature measurement. A Pt100 is 100 Ω at 0 °C and changes about 0.39 Ω/°C near 0 °C.

The tolerance classes are tight. Joseph Wu’s RTD guide for TI lists IEC 60751 class A as ±(0.15 + 0.002|T|) °C and class B as ±(0.3 + 0.005|T|) °C, which is ±0.35 °C and ±0.8 °C respectively at 100 °C.

The errors that matter in practice are around the sensor, not in it.

Lead resistance. In a two-wire connection the lead resistance adds directly to the RTD. As an illustrative example, 10 m of 24 AWG copper is about 0.84 Ω per conductor, so the two leads add 1.68 Ω. Divided by 0.385 Ω/°C, that’s a 4.4 °C error. A three-wire connection cancels the leads if they match; a four-wire connection removes them from the measurement entirely, which the TI guide describes as the most accurate configuration.

Self-heating. The excitation current heats the element. The TI guide gives the error as ΔT = I²R / E, where E is the self-heating coefficient, about 2.5 mW/°C for small thin-film elements and up to about 65 mW/°C for larger wire-wound ones. At 1 mA, a Pt100 at 0 °C dissipates 0.1 mW, which is about 0.04 °C in a small thin-film element. The guide recommends keeping most excitation currents below 1 mA.

Thermistors: lots of signal over a narrow range

An NTC thermistor is a ceramic semiconductor whose resistance falls steeply as temperature rises. The sensitivity is the attraction. Using a common model with an illustrative B value of 3,950 K, a 10 kΩ thermistor changes about −4.4 %/°C at 25 °C, which is about 444 Ω/°C, compared with 0.39 Ω/°C for a Pt100. A simple divider into a microcontroller ADC can resolve small temperature changes.

The price is nonlinearity. With the same illustrative model, the resistance is about 33.6 kΩ at 0 °C and 1.09 kΩ at 85 °C, and the sensitivity falls to about −3.1 %/°C at 85 °C. A straight-line conversion is only good over a few degrees. The standard fix comes from oceanography: in 1968, John Steinhart and Stanley Hart published in Deep-Sea Research an empirical equation, 1/T = A + B ln R + C (ln R)³, that fits thermistor curves very closely with three calibration points. It’s still the default in thermistor firmware.

Thermistors are also limited in range (typically somewhere within −50 to 150 °C, depending on the part) and, like RTDs, they self-heat. Their high resistance makes lead resistance less important than for an RTD.

Thermocouple vs RTD vs thermistor compared

Property	Thermocouple (type K)	RTD (Pt100)	NTC thermistor (10 kΩ, illustrative)
Specified range	−270 to 1,370 °C	Typically −200 to 850 °C (by class and construction)	Typically within −50 to 150 °C
Signal	≈39.5 µV/°C	≈0.39 Ω/°C	≈−4.4 %/°C at 25 °C
Standard tolerance example	Class 2: ±3.75 °C at 500 °C	Class B: ±0.8 °C at 100 °C	Set by the part’s datasheet
Linearity	Moderate; reference functions needed	Near-linear; Callendar–Van Dusen	Strongly nonlinear; Steinhart–Hart
Main hidden error	Cold-junction compensation	Lead resistance, self-heating	Nonlinearity, self-heating
Excitation	None (self-generating)	Current source, ideally ≤1 mA	Divider or current source
Typical uses	Furnaces, exhausts, kilns	Process control, labs, pharmaceuticals	Batteries, HVAC, consumer devices

What to actually do

Choose by range first

If the measurement goes above about 600 °C, start with a thermocouple. If it stays within roughly −50 to 150 °C and cost or simplicity matter, start with a thermistor. In between, or when accuracy and stability dominate, start with an RTD.

Write the error budget, including the hidden term

For a thermocouple, include the sensor tolerance and the cold-junction measurement. For an RTD, include the class tolerance, lead resistance for your wiring and self-heating. For a thermistor, include the interpolation error of your conversion and self-heating.

Wire for the sensor

Use four-wire (or matched three-wire) connections for RTDs over any real distance. Use thermocouple extension wire of the right type for thermocouples, and keep the cold-junction sensor at the same temperature as the terminals.

Convert correctly in firmware

Use the NIST/IEC reference functions for thermocouples (with voltage-domain cold-junction compensation), Callendar–Van Dusen for RTDs and Steinhart–Hart or the manufacturer’s table for thermistors.

What not to do

Don’t add the cold-junction temperature to the thermocouple temperature; add voltages. And don’t run a Pt100 at several milliamps to get more signal without calculating the self-heating error.

How to put it to work

If you have an hour

For your current design, write the range, the required accuracy and the cable length, and calculate the one hidden error for your sensor type: cold-junction uncertainty, lead resistance or linearization error.

If you have a sprint

Put a reference thermometer next to your sensor in a stirred bath or dry block at three temperatures across the range, and compare. The residuals show whether the error comes from the sensor class, the wiring or the conversion.

If you want it to last the product’s lifetime

Record the sensor type, class, wiring and conversion method in the design documentation, and recheck the reference comparison when any of them changes. Thermocouples in particular drift with exposure to high temperature, so plan a replacement or recalibration interval.

The question to ask yourself: “Which error that isn’t on the sensor’s datasheet is the largest in my system, and have I measured it?”

Frequently asked questions

What is the difference between a thermocouple, an RTD and a thermistor?

A thermocouple generates a small voltage from the temperature difference between two junctions of dissimilar metals. An RTD is a platinum resistor whose resistance rises nearly linearly with temperature. A thermistor is a semiconductor resistor whose resistance changes steeply and nonlinearly, usually falling as temperature rises.

Which temperature sensor is most accurate?

Of the three, platinum RTDs usually offer the best accuracy and stability over a wide range; IEC 60751 class A is ±0.35 °C at 100 °C. Calibrated thermistors can be very accurate over a narrow range. Thermocouples have the widest range but the loosest standard tolerances.

Why do thermocouples need cold-junction compensation?

A thermocouple’s voltage depends on the temperature difference between the measuring junction and the point where it connects to the electronics. The electronics must measure that reference temperature and compensate, preferably by converting it to an equivalent voltage before converting the total to temperature.

Should I use a 2-wire, 3-wire or 4-wire RTD?

Use 2-wire only for very short leads or low accuracy needs, since lead resistance adds directly to the reading. Three-wire cancels lead resistance if the leads match, and four-wire removes it from the measurement and is the most accurate.

What is the Steinhart–Hart equation?

It’s an empirical equation, 1/T = A + B ln R + C (ln R)³, published by Steinhart and Hart in 1968, that relates a thermistor’s resistance to absolute temperature. The three coefficients are found by calibrating at three temperatures.

Related reading: sensor calibration: offset, gain and linearity, how to read a sensor datasheet and grounding and shielding sensor signals.

Sources

1. Steinhart, J. S., & Hart, S. R. (1968). “Calibration Curves for Thermistors.” Deep-Sea Research and Oceanographic Abstracts, 15(4), 497–503. https://doi.org/10.1016/0011-7471(68)90057-0
2. Callendar, H. L. (1887). “On the Practical Measurement of Temperature: Experiments Made at the Cavendish Laboratory, Cambridge.” Philosophical Transactions of the Royal Society of London A, 178, 161–230. https://doi.org/10.1098/rsta.1887.0006
3. Burns, G. W., Scroger, M. G., Strouse, G. F., Croarkin, M. C., & Guthrie, W. F. (1993). Temperature-Electromotive Force Reference Functions and Tables for the Letter-Designated Thermocouple Types Based on the ITS-90. NIST Monograph 175. https://doi.org/10.6028/NIST.MONO.175
4. International Electrotechnical Commission. IEC 60751, Industrial platinum resistance thermometers and platinum temperature sensors; IEC 60584-1 and 60584-2, Thermocouples.
5. Wu, J. (2018, rev. 2023). “A Basic Guide to Thermocouple Measurements.” Texas Instruments Application Report SBAA274A. https://www.ti.com/lit/an/sbaa274a/sbaa274a.pdf
6. Wu, J. (2018, rev. 2023). “A Basic Guide to RTD Measurements.” Texas Instruments Application Report SBAA275A. https://www.ti.com/lit/an/sbaa275a/sbaa275a.pdf