Two instruments can both claim 0.5 percent accuracy and be wildly different where it counts. The trick is in the three little words after the number: percent of reading, percent of span, or percent of full scale. A percent-of-span spec fixes the error in absolute terms across the whole range, which means it becomes a huge fraction of a small reading near the bottom of the range - and that is the hidden reason turndown claims quietly collapse. This page explains the three ways vendors quote accuracy, why the difference explodes at low process values, and how to convert a datasheet number into a real engineering-unit error band you can trust at 10 percent of range.
% of Reading vs % of Span in one line: Percent-of-reading accuracy expresses error as a percentage of the actual value being measured, so the absolute error shrinks as the reading drops. Percent-of-span (or percent-of-full-scale) accuracy expresses error as a percentage of the whole measuring range, so the absolute error is fixed regardless of the reading. Near the bottom of a range, a percent-of-span device is far less accurate than a percent-of-reading device with the same headline number, which is why low-flow and high-turndown accuracy claims must be read carefully.
A percent-of-reading spec ties the error to the value you are actually measuring. If a flow meter is quoted at 0.5 percent of reading and it is measuring 100 units, the error is half a unit; if it is measuring 20 units, the error is a tenth of a unit. The absolute error scales down with the reading, so the meter stays proportionally just as good at the low end as at the high end. This is the most demanding and most desirable form of spec, and it is what you want on anything that operates over a wide range.
A percent-of-span spec ties the error to the entire calibrated range instead. Span is the distance from the bottom of the range to the top - if a transmitter is ranged 0 to 200 units, its span is 200. A 0.5 percent of span spec means the error is 0.5 percent of 200, which is one unit, and that one unit is fixed everywhere in the range. At a reading of 100 it is a one percent error; at a reading of 20 it is a five percent error; at a reading of 10 it is a ten percent error. Same instrument, same absolute error, but the percentage relative to the reading balloons as the reading falls.
Percent of full scale is essentially the same idea as percent of span whenever the range starts at zero, tying the error to the upper range value. The three phrasings look almost interchangeable on a glossy datasheet, but percent of reading behaves completely differently from the other two across the range. Whenever you see an accuracy figure, the units after it - of reading versus of span or full scale - matter as much as the number itself, and a spec quoted with no reference at all should be treated with suspicion.
The gap between the two spec types is invisible at the top of the range and enormous at the bottom, and that asymmetry is exactly where it bites. At full scale, 0.5 percent of reading and 0.5 percent of span are identical, because the reading equals the span. As the process value drops, the percent-of-reading error drops with it while the percent-of-span error stays frozen at its full-scale value, so the two curves fan apart. By the time you are operating at 10 percent of range, the percent-of-span device's fixed error is ten times larger as a fraction of the reading than it was at full scale.
This is the mechanism behind collapsing turndown claims. Turndown - the ratio of the highest to the lowest flow a meter can measure usefully - sounds like a property of the meter, but usable turndown is really limited by how much error you can tolerate at the low end. A meter with a headline percent-of-span accuracy might advertise a wide turndown, but at the bottom of that range the fixed span-based error can swamp the reading, so the low-flow measurement is nearly worthless even though the meter is technically still reading. The turndown was real on paper and fictional in engineering units.
Percent-of-reading instruments hold their turndown far better, because their error shrinks alongside the flow, keeping the proportional accuracy roughly constant down into the low end. This is why the reference basis of an accuracy spec is one of the first things to check when a measurement must be trusted across a wide operating range, such as a well that flows hard early and dwindles later, or a process that turns down overnight. Two meters with the same accuracy headline can be a factor of ten apart at the flows you actually care about.
The way to defend against a misleading headline is to translate every spec into an absolute engineering-unit error at the flows you actually run, then decide if it is acceptable. For a percent-of-span or full-scale spec, take the percentage times the span, and that single absolute number is your error band everywhere - the same plus-or-minus in engineering units at 10 percent as at 100 percent. For a percent-of-reading spec, take the percentage times the specific reading of interest, giving a smaller band at low readings. Doing this at your lowest normal operating point, not just at full scale, is what exposes the real difference between two candidate instruments.
A worked feel for it: a device ranged 0 to 200 units quoted at 0.5 percent of span has a fixed error of one unit. At a 10 percent reading of 20 units, that one unit is a five percent error in engineering terms. A different device on the same service quoted at 0.5 percent of reading has, at that same 20-unit reading, an error of a tenth of a unit - fifty times tighter at the flow that matters. The datasheets both said 0.5 percent; the operating reality at low flow is not remotely comparable. Some vendors publish a combined spec, such as a percent-of-reading term plus a small percent-of-span floor, and those should be evaluated the same way at the low end where the floor dominates.
This conversion is exactly the kind of context that belongs in the SCADA layer, because the operator reading a value on a dashboard has no way to know how much to trust it otherwise. When a cloud SCADA platform such as Merobix historizes a flow, documenting the meter's real engineering-unit error band - especially at low flow - lets an engineer set alarm limits and interpret custody or allocation figures with the actual uncertainty in view. A reading of 20 units that is really 20 plus or minus one means something very different from 20 plus or minus a tenth, and capturing that alongside the trend keeps a thin, uncertain low-flow measurement from being read with false confidence.
For measurements that operate over a wide range, percent of reading is far better, because its absolute error shrinks as the reading falls, keeping accuracy proportional at low values. Percent of span holds the error fixed across the whole range, so it becomes a large fraction of a small reading near the bottom. If you only ever operate near full scale, the two are similar; if you turn down, percent of reading wins decisively.
Usable turndown is limited by how much error you can tolerate at the lowest flow, and a percent-of-span accuracy spec keeps its absolute error fixed while the flow drops. At the bottom of the range that fixed error becomes a huge fraction of the reading, so the low-flow measurement is unreliable even though the meter is still technically reading. The advertised turndown was real on paper but not in engineering units.
For percent of span or full scale, multiply the percentage by the span to get a fixed absolute error that applies everywhere, including at 10 percent. For percent of reading, multiply the percentage by the actual reading, which gives a smaller error at low readings. Comparing candidate instruments at your lowest normal operating point, in engineering units, is what reveals the real difference between two specs with the same headline number.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.