I have a multimeter with the following temperature accuracy for its Type K probe

±3% of reading ±5C

The display is in steps of 1C

I attempted some calibration:

  • In ice water, the reading is 0C
  • In boiling water, the reading is 94C (1012.9mbar/100C)

Can I interpolate that? Would the reading at 20C be 0.2 x 94C = 18.8 C, i.e. "19" or maybe even "18"? Or are there other factors at play?

I would expect the readings to be monotonous against the actual values, but perhaps the ADC's step size and non-linearity, in combination with the display's granularity, make readings not strictly monotonous, leaving scattered "pleateaus" of readings.

Of course, one could mix the 0C water and 100C at various ratios to create a actual vs. reading chart, but not only is this tedious, I am sure also fraught with systematic and random errors. So I'd rather go with the wisdom of the group here.

The model is "DMR-6500", but perhaps the validity of your reasoning will be independent from the model/brand.

  • \$\begingroup\$ Interpolating like that could reduce your accuracy at 20C than if you just took it at face value, could increase it for readings closer to 100C. Also, boiling temperature of water is atmospheric pressure dependent. So there's that. What's the elevation of where you are at? if 94C for boiling H2O is accurate then you are at 6000 feet. How accurate do you actually need this? Past a certain point you should just buy a proper thermometer instead of using the after-though feature on your DMM. \$\endgroup\$
    – DKNguyen
    Jul 29, 2020 at 19:37
  • \$\begingroup\$ @DKNguyen It's 29.91 (1012.9mbar) here today, so 100C. Good point. Edited question. I am looking for 0.5C accuracy. Yes, better thermometer is the right solution, but for once I wanted to make use of an "after-thought"! \$\endgroup\$
    – P2000
    Jul 29, 2020 at 20:02
  • \$\begingroup\$ you have taken two points, why not do a 2-point calibration to your data? \$\endgroup\$
    – Aaron
    Jul 29, 2020 at 20:05
  • \$\begingroup\$ @Aaron, yes that's what the 20C example calculation was (or was I not clear?). \$\endgroup\$
    – P2000
    Jul 29, 2020 at 20:07
  • \$\begingroup\$ try asking at chemistry.stackexchange.com \$\endgroup\$
    – jsotola
    Jul 29, 2020 at 23:56

3 Answers 3


There's a decent chance your multimeter does not bother to linearize the thermocouple curve. The uV/°C is less between 0°C and 100°C (40.96uV/K) than between 500 and 600°C (42.61uV/K) so you'd expect some sag at the 100°C point if they just average out the errors over the 760°C range.

Most older design multimeters are very linear (can't speak for yours), but some newer ones use different ADC methods and are not impressive. So if it's very linear, 'calibrating' it may work, but only if the ambient temperature (not the temperature being measured) is very stable and close to the calibration conditions. The cold-junction compensation of the multimeter is probably not too impressive. The internal temperature sensor error contributes approximately a 1:1 error to the reading.

Frankly, in the 0..100°C range you'd probably be better off buying a precision NTC thermistor or 1K RTD and measuring the resistance on the DMM.

  • 1
    \$\begingroup\$ Yes, this was my concern too: "but only if the ambient temperature (not the temperature being measured) is very stable and close to the calibration conditions". Let me look into the NTC/RTD. \$\endgroup\$
    – P2000
    Jul 29, 2020 at 20:51

If the type K probe is made properly you ought to get good results to +/- 0.5'C.

Like any ADC calibration, you test for errors in offset, gain, monotonicity, linearity and dead-bands and if desired the RMS integrated average squared error (IASE) for every value.

Find the value that is not affected by gain or temperature of the sensor, perhaps 0'C then null offset error. Knowing altitude does not affect 0'C but affects boiling point -0.5'C/500ft you can use ice water and boiling water to get < 0.1'C error reference points.

  1. null the error for 0'C
  2. null the gain error for 100'C
  3. test for monotonicity with a scope and D/A converter output that has been calibrated. Dead-codes or hysteresis are a sign of digital to analog crosstalk on the ground path at binary transitions like xxxx1111 to xxx1000. If the apparent signal or Vref shifts with the binary state, you get some missing or "dead codes".

Do you have a 12 to 16 bit ADC for adequate resolution and accuracy of 200 readings with 0.5'C increments or a carefully scaled 8 bit with 1/2 bit error in 256 deg C range or perhaps only 10 to 40'C (?)

For linearity and accuracy, I prefer using transistors as diodes Vbe=Vce for measuring temp. over thermistors and get 0.3'C accuracy with some effort. But now the LM75 is pretty useful.

Maybe useful info

This is all anecdotal info.

Your DMM specs are pretty loose so you can correct the gain error with a calculator.

If you want to compare a 1N4448 diode to say a PN2222A with BC shorted, try the diode test that supplies on your meter 300uA and has 1mV resolution.

Depending on diode leakage you may get between 300mV and 500mV @ 25'C then test 0'C and 100'C to compare. With a diode tempco of -2.3 mV/'C that is almost your 0.5'C resolution requirement with a good low leakage transistor.

  • \$\begingroup\$ Your "test for monotonicity" (and dead codes) focuses entirely on ADC and display. You are replacing the probe with something else, like a controlled voltage source (hence your DAC comment)? Maybe also a power supply and divider? \$\endgroup\$
    – P2000
    Jul 29, 2020 at 20:55
  • \$\begingroup\$ Yes , you ought never to assume your ADC is perfect. Analog DAC comparison is a great way to quickly compare with a thermal sweep from 0 to 100 to 0 to check for missing codes and linearity on a DSO. \$\endgroup\$ Jul 29, 2020 at 20:59

I have a 'DMM with K-type thermocouple input', and a K-type digital thermometer.

The DMM reads reasonably accurately over the 15-30C 'lab ambient' range of temperatures, with increasing error above that, until it's many many degrees out at 100 C.

The digital thermometer, in the same size case and not costing much more, is quite accurate at 0C, 100C, and all temperatures in between.

A low cost add-on to a DMM, especially if it uses the 4 mm sockets rather than a dedicated K-type connector, is not a 'thermometer', it's a sales gimick.

It would be worth establishing whether your DMM linearises the K-type curve or not. Compare it with a mercury in glass thermometer at several temperatures, and then compare that with the linearity curve for the K-type. If you're going to try to improve the accuracy of the reading between reference temperatures, then you need to know whether to interpolate a line, or to fit a curve.

  • \$\begingroup\$ Yes, I won't argue that this is likely a sales feature and I would use it maybe to determine whether a transistor or IC is hot or overheating. The DMM's thermal probe has a two-prong plug for the 4mm sockets, and the mode must be set to "C" (or "F"). Not sure what's in the plug, but I suspect any linearization would be done in the DMM. Why would the type of socket preclude it from being accurate? Cheap feature = cheap performance? \$\endgroup\$
    – P2000
    Jul 30, 2020 at 4:00
  • \$\begingroup\$ A K-type plug/socket is made from the same K-type metals, so that the temperature of the plug doesn't matter. The 4 mm sockets on the meter are whatever they happen to be made of, so the temperature of those will introduce errors into the reading. \$\endgroup\$
    – Neil_UK
    Jul 30, 2020 at 4:56
  • \$\begingroup\$ Aha, and linearization in a DMM can't make up for that. \$\endgroup\$
    – P2000
    Jul 30, 2020 at 5:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.