You're troubleshooting a temperature sensor. Day to day, the datasheet says "resistance-based. Is that right? 7 ohms at room temperature. Is it broken? On top of that, " Your multimeter reads 109. Or did you just buy the wrong thing?
Here's the short answer: RTDs and thermistors both use resistance to measure temperature. But they behave differently, cost differently, and live in completely different worlds Took long enough..
If you've ever wondered why your 3D printer uses a thermistor but the lab down the hall swears by platinum RTDs — this is the article for you Most people skip this — try not to. And it works..
What Is a Resistance Temperature Detector
An RTD — Resistance Temperature Detector — is exactly what it sounds like. Which means a sensor whose electrical resistance changes predictably with temperature. In practice, no junctions. No thermocouple voltages. Just a length of metal wire (usually platinum) that gets more resistive as it heats up No workaround needed..
The most common industrial standard? **Pt100.Worth adding: ** That means 100 ohms at 0°C. There's also Pt1000 (1000 ohms at 0°C), and less common nickel or copper versions Simple, but easy to overlook. Turns out it matters..
Platinum wins because it's stable. And its resistance-temperature curve is repeatable across decades. That's why you'll find Pt100s in pharmaceutical manufacturing, aerospace test stands, and food processing plants where a 0.It doesn't oxidize much. 1°C drift matters.
How the physics actually works
Metals conduct electricity via free electrons. Consider this: resistance goes up. As temperature rises, the metal lattice vibrates more. Electrons scatter more. It's nearly linear over wide ranges — especially platinum.
The Callendar-Van Dusen equation models it precisely. But in practice? Most people use a lookup table or a transmitter that handles the math And that's really what it comes down to..
RTD wiring configurations matter
Two-wire RTDs exist. They're cheap. They're also inaccurate because lead resistance adds directly to your reading. Don't use them for anything you care about The details matter here..
Three-wire is the industrial standard. Two leads carry current, one measures voltage drop. The transmitter assumes all three leads are identical — close enough for most factory runs.
Four-wire? That's lab grade. Current on two wires, voltage sense on two separate wires. True Kelvin sensing. Lead resistance drops out completely. If you're calibrating other sensors, this is what you use Surprisingly effective..
What Is a Thermistor
Thermistor = thermal resistor. In real terms, same core idea — resistance changes with temperature — but the material is a semiconductor, not a metal. Usually a metal oxide ceramic (manganese, nickel, cobalt oxides) It's one of those things that adds up..
Two flavors. NTC (Negative Temperature Coefficient): resistance drops as temperature rises. PTC (Positive Temperature Coefficient): resistance rises. NTC dominates temperature sensing. PTC shows up in self-resetting fuses and motor protection And it works..
A typical 10k NTC thermistor reads ~10,000 ohms at 25°C. At 100°C? Maybe 600 ohms. That's a massive swing — great for resolution, terrible for linearity.
The Steinhart-Hart equation
You don't linearize a thermistor with a straight line. You use Steinhart-Hart:
1/T = A + B·ln(R) + C·(ln(R))³
Three coefficients. Most datasheets give you A, B, C. Day to day, or just a resistance-vs-temperature table. Your microcontroller firmware handles the rest.
Why It Matters: Choosing Wrong Costs You
I've seen engineers spec a Pt100 for a battery pack monitor. Overkill. Needs a precision current source and 4-wire routing. Also, expensive. Now, a 10k NTC with a voltage divider and an ADC pin? Done in five minutes for fifty cents.
Flip side: a food safety auditor walks your line. 5°C over six months. Because of that, you fail. But your thermistor-based loggers drift 0. Should've used Class A Pt100s with calibrated transmitters And it works..
The device you pick determines:
- Accuracy ceiling — RTDs win long-term
- Sensitivity — thermistors win at low temps
- Wiring complexity — thermistors are simpler
- Cost per node — thermistors win below $5
- Interchangeability — RTDs are standardized; thermistors vary by vendor
How They Work: Side by Side
RTD operation
Constant current (typically 1 mA) flows through the element. Voltage drop = I × R(T). Measure voltage, calculate R, convert to T.
Self-heating is real. In still air, that's maybe 0.Consider this: in vacuum? But significant. And 05°C error. 1 mA through 100 ohms = 0.1 mW. Which means in flowing water? Negligible. Pulse the current if you're paranoid Not complicated — just consistent..
Thermistor operation
Voltage divider. Fixed resistor + thermistor to ground. ADC reads midpoint. As T changes, R_therm changes, V_out changes Most people skip this — try not to..
Ratiometric measurement (using the same reference for ADC and divider top) cancels supply noise. Think about it: do this. Always.
Signal conditioning
RTDs need:
- Precision current source or ratiometric excitation
- High-resolution ADC (24-bit delta-sigma typical)
- Lead compensation (3 or 4 wire)
- Linearization (Callendar-Van Dusen or polynomial)
Thermistors need:
- Voltage divider resistor (pick for max sensitivity at your target temp)
- ADC (12-bit often enough)
- Steinhart-Hart or lookup table in firmware
- Maybe a series resistor to limit self-heating current
Common Mistakes / What Most People Get Wrong
Mistake 1: Assuming all Pt100s are interchangeable. Class B: ±0.3°C at 0°C. Class A: ±0.15°C. Class AA (1/3 DIN): ±0.1°C. Tolerance stacks up fast. If you swap a Class B for Class A without recalibrating, your data is suspect.
Mistake 2: Using a 2-wire RTD with 10 feet of cable. 24 AWG copper is ~0.025 ohms/foot. 20 feet round trip = 0.5 ohms = 1.3°C error on a Pt100. That's not noise. That's a systematic offset you'll never debug unless you know to look Took long enough..
Mistake 3: Picking a thermistor divider resistor at random. The divider resistor sets your sensitivity curve. Match it to the thermistor's resistance at your target temperature, not 25°C. If you're monitoring 80°C oil, pick R_divider ≈ R_therm(80°C). Otherwise you waste ADC codes.
Mistake 4: Ignoring self-heating in still air. A 10k NTC with 5V across a 10k divider? 1.25 mW at 25°C. In free air, that's 1–2°C of self-heating. Pulse the measurement. Or use a higher divider resistor. Or accept the offset and calibrate it out.
Mistake 5: Mixing thermistor models in the same system. Vendor A's "10k NTC" has different Steinhart-Hart coefficients than Vendor B's. They're not drop-in replacements. Lock down a part number. Qualify it. Treat it like a calibrated component Nothing fancy..
Practical Tips / What Actually Works
For RTDs:
- Use 3-wire minimum. 4-w
-wire if the budget allows and the run is long. The fourth wire eliminates lead resistance entirely; the third gets you most of the way there for half the connector cost Still holds up..
- Excite with 1 mA or less unless the application demands otherwise. Lower current means less self-heating, and you can always gain up the ADC if the signal is small.
- Burn-in new sensors for 24–48 hours before calibration. Drift in the first day is real, especially on thin-film elements.
- Keep the current source outside the thermal gradient zone. A current source that sits next to the heater will drift with temperature and take your reference with it.
For thermistors:
- Buy the beta or Steinhart-Hart coefficients from the vendor's datasheet lot, not just the generic "10k at 25°C" line. If they won't give you lot-specific data, find a vendor who will.
- Place the divider resistor physically close to the ADC, not close to the sensor. The sensor lead is the noisy part; the resistor is the stable part.
- Use a pulldown to ground rather than pullup to VCC if your ADC reference is ground-referenced. It simplifies the ratiometric math and avoids a subtracted offset in firmware.
- If you need fast response, use a bead-in-glass or chip-on-board style. If you need chemical resistance, use a stainless probe even though it lags.
Cross-cutting:
- Log the raw ADC counts, not just the computed temperature. When a sensor drifts or a coefficient is wrong, the raw data lets you reconstruct what actually happened.
- Validate against a calibrated reference at three points: cold, ambient, and hot. One-point "it reads 25°C in the office" checks catch nothing.
- Document the excitation scheme, resistor values, and firmware equation in the repo. Six months from now, nobody will remember whether R_divider was 9.76k or 10.2k.
Conclusion
RTDs and thermistors are both mature, cheap, and capable—but they fail in different ways and reward different disciplines. Pick based on range, accuracy requirement, cable length, and how much firmware math you're willing to own. And neither is "better" in the abstract. Plus, rTDs demand lead management, stable excitation, and respect for tolerance classes; thermistors demand coefficient discipline, divider design, and self-heating awareness. Then lock the part number, document the circuit, and validate against a reference before you trust a single degree The details matter here..
