As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
They describe different things, and both can be true at once. Resolution is the size of the step the display can show, so a tenth of a decibel resolution means the reading moves in increments of a tenth. Accuracy is how far the whole reading may sit from the true sound pressure level, so two decibels means the real level could be anywhere in a band four decibels wide centred on what you see. A meter reading 86.3 dB with those two specifications is telling you that the true level is somewhere between 84.3 and 88.3, and that it can resolve a change of a tenth within that uncertain band. The fine display is for comparing, the tolerance governs what you may claim absolutely.
Resolution is the step the display can show, accuracy is how far it can be wrong
Every measuring instrument has these two properties and they are independent of each other. Resolution is a property of the output stage: the number of digits, the size of the least significant one, and the quantisation of the converter behind it. An instrument can be given more digits cheaply, and more digits make it look more precise without making it more correct.
Accuracy is a property of the whole chain, from the physical transducer through the analogue conditioning to the arithmetic. For a sound level meter the chain begins at a microphone whose sensitivity varies between individual units, varies with temperature and atmospheric pressure, and varies with frequency in a way that is never perfectly flat. Then comes the weighting filter, which is an electronic approximation to a defined curve, the detector that forms a time average, and the logarithmic conversion to decibels. Each stage contributes.
Two of the listings here state both figures explicitly and identically. The Extech 407730-NIST sound level meter states two decibel accuracy with a tenth of a decibel resolution, A and C weighting, an analogue output and recording of maximum and minimum values over time. The Extech 407730 states the same pairing with a range of forty to one hundred and thirty decibels and a fast responding analogue bar graph. Stating both is the honest presentation, and it is less common than it should be.
Where the 2 dB tolerance on a sound level meter comes from
A tolerance figure is a budget, and the items in the budget are physical. The largest single contributor is usually microphone sensitivity, since every capsule differs slightly from nominal and the difference is temperature and pressure dependent. The frequency response of the capsule adds more, because a diaphragm cannot be equally sensitive at thirty hertz and at ten kilohertz, and the deviation grows at both ends.
The case contributes through diffraction. Sound arriving from different directions bends around the body of the instrument differently depending on wavelength, so the level at the diaphragm depends on orientation in a way that becomes significant once the wavelength approaches the size of the case. This is why a measurement standard specifies how a meter should be held and pointed, and why a reading taken with the instrument flat on a bench is not the same reading as one taken held at arm length.
The weighting filter adds its own tolerance. A weighting is defined as a curve of gain against frequency, and a real filter approximates that curve within permitted limits. The permitted limits are narrow around one kilohertz, which is the reference frequency where the weighting passes through zero gain, and they widen considerably at the extremes where the curve is steep. Published acceptance limits for instrument classes reflect this directly: at the reference frequency they are tight, commonly quoted as about one decibel for the stricter class and a little under one and a half for the looser one, and they widen to several decibels at the edges of the range. A single headline figure such as two decibels is a summary of that envelope rather than a constant.
Worked example: reporting a reading with its uncertainty attached
Suppose you are checking levels against an action threshold of eighty five decibels and the meter states two decibel accuracy. Each reading becomes a band rather than a point, and the question is which bands actually settle the matter.
| Displayed reading | Band the true level may occupy | Position relative to an 85 dB threshold | What you may state |
|---|---|---|---|
| 80.4 dB | 78.4 to 82.4 dB | Entirely below | Below the threshold, on this instrument, at this position |
| 84.6 dB | 82.6 to 86.6 dB | Straddles | Undetermined, the instrument cannot settle it |
| 86.3 dB | 84.3 to 88.3 dB | Straddles | Undetermined, the instrument cannot settle it |
| 88.9 dB | 86.9 to 90.9 dB | Entirely above | Above the threshold, on this instrument, at this position |
The two middle rows are the uncomfortable ones and they are also the common ones, because thresholds tend to be set where real situations cluster. With a two decibel tolerance there is a band four decibels wide in which the instrument genuinely cannot tell you which side of the line you are on, and no amount of staring at the tenths digit will resolve it. That is not a defect. It is the specification doing exactly what it says, and the only honest responses are to use an instrument with a tighter tolerance, to treat the case as exceeding until shown otherwise, or to record the ambiguity.
Notice also what happens to the arithmetic of differences. If the same meter reads 84.6 at one position and 88.9 at another, the difference of 4.3 decibels is far better determined than either absolute value, because the microphone sensitivity offset and most of the filter error are the same in both readings and subtract out. The tolerance that applies to a difference measured on one instrument within a short interval is much smaller than the tolerance on either figure.
Why the last digit still earns its place
Everything above explains why the tenths digit is not decoration. Comparative work dominates real use: is this machine louder than that one, did the enclosure help, is the level at the boundary higher now than last month, which of these two positions is worse. All of those are differences taken on one instrument, and all of them benefit from a display that can show a change of a tenth.
The tenths digit also makes the behaviour of the detector visible. Watch a reading settle on fast weighting and you see it flicker across several decibels; switch to slow and the same source produces a steadier figure. That difference is information about the source, and a display quantised to whole decibels would hide small fluctuations entirely while exaggerating the appearance of large ones as the reading jumped between steps.
A tighter tolerance comes from the instrument class rather than from the display, and the two are often confused. The CRYSOUND CRY2831 states conformity with IEC 61672-1 Class 2, a measurement range of twenty nine to one hundred and thirty eight decibels, a frequency range of twenty hertz to twelve and a half kilohertz, A, C and Z weighting, fast and slow time weighting, an IP54 rating and a calibration certificate. A numeric tolerance in decibels is not stated in the listing bullets, and neither is a display resolution, which is the mirror image of how the two Extech listings present themselves. Between them the three show the two halves of the specification rarely appearing together.
Writing down a level so the number survives a challenge
A sound measurement is often written down because somebody disagrees about it, so write it in a form that answers the obvious questions before they are asked. Record the level, the frequency weighting, the time weighting, the duration over which it was observed and whether the figure is an instantaneous reading, a maximum or an average. Record the position, including distance from the source and height above the floor, and the orientation of the microphone. Record whether a windshield was fitted. Record the instrument, its stated tolerance and when it was last checked against a reference.
Then state the conclusion in terms the tolerance supports. If the band straddles the threshold, say so rather than rounding towards the answer you expected. A figure written as a level with its tolerance, its weighting and its conditions is a measurement. The same figure written alone is an assertion, and an assertion with a tenths digit on it is simply a confident assertion. The relationship between the three specifications that get muddled here is worked through in how to compare sensor accuracy, resolution and repeatability, the instrument itself is introduced in what a sound level meter is, and the selection questions that follow from the tolerance you need are covered in how to choose a sound level meter and the sound level meter overview.
Frequently Asked Questions
Is a display showing tenths of a decibel pointless?
No, because the tenths carry information about differences even when they carry none about absolute level. The large part of a tolerance figure is systematic, meaning it shifts every reading from that instrument in the same direction by roughly the same amount. When you subtract two readings taken moments apart on one meter, that shared offset cancels and the small difference that remains is far better determined than either figure on its own. Coarse display steps would throw that information away.
Does a calibration certificate shrink the tolerance?
It replaces a generic tolerance with a specific one, which is usually an improvement. A certificate records what the instrument read against a reference at stated frequencies and levels, with an uncertainty for the comparison itself. If the certificate shows a consistent offset you can correct for it, and the remaining uncertainty is the certificate uncertainty plus whatever has drifted since. What it does not do is make the instrument conform to a tighter class, because class is about behaviour across the whole frequency range and not only at the calibration point.
How should a reading be written down?
As a value, a tolerance, a weighting and a time response, in that order, with the duration beside it. A bare figure invites a reader to treat it as exact. Writing a level together with the stated instrument tolerance, the frequency weighting used and whether the detector was set to fast or slow makes the measurement reconstructible and makes its limits visible at a glance. A measurement that cannot be reconstructed cannot be defended.
Why does accuracy depend on frequency?
Because the microphone, the case and the electronics all behave differently across the spectrum. A microphone diaphragm has a natural resonance and a size that becomes comparable to the wavelength at high frequencies, so its response rises or falls away from flat at the extremes. The instrument case diffracts sound around itself in a frequency dependent way. The weighting filter is an approximation to a curve, and the tolerance on that approximation widens where the curve is steepest. A single tolerance figure quoted at one frequency hides all of this.
Can I average several readings to beat the tolerance?
Averaging reduces random variation and does nothing at all to systematic error. If the instrument reads consistently high because of a microphone sensitivity offset, a hundred readings average to the same wrong answer with a tighter spread around it, which is worse than useless because the tighter spread looks like improved confidence. Averaging helps with a fluctuating source and with reading noise. Only a comparison against a reference addresses the systematic part.