Guide
Why do surface roughness measurements disagree between supplier and customer, and what should the drawing specify?
Surface roughness measurement is the process of tracing or scanning a surface, filtering the recorded profile and calculating a parameter such as Ra from it, so the number it returns depends on the measurement settings as much as on the surface itself. When a supplier and a customer read different values on the same part, the cause is usually a difference in cutoff, filter, evaluation length, trace direction, location or acceptance rule, and the cure is a drawing callout that fixes those settings.
By Saola Manufacturing engineering team · Published · Updated
On this page
How do you tell a measurement disagreement from a bad part?
If two instruments disagree on the same marked spot of the same part, the problem is the measurement method, not the part. If they agree on the method and still disagree, or if the readings scatter widely across one surface, look at the instruments and at the surface itself.
A typical dispute looks like this: the supplier’s inspection report shows an Ra of 0.8 µm on a sealing face, the customer’s receiving lab reads an Ra of 1.4 µm on the same face, and the lot is put on hold. Neither lab is necessarily wrong. Ra is the arithmetic mean of the absolute profile heights after filtering, and every step before that arithmetic, from where the stylus is placed to which wavelengths the filter keeps, is a choice. The standards for profile roughness, now the ISO 21920 series in the ISO system and ASME B46.1 in the US, define defaults for those choices, but a default only applies when the drawing is silent and both labs actually use it.
Before arguing about who is right, collect three things from each side: the full instrument printout or report (not just the number), the settings used (cutoff, short-wavelength filter, evaluation length, stylus tip, skid or skidless pickup) and a sketch or photo of where and in which direction the trace was taken. In most disputes the answer is visible as soon as the two sets of settings sit side by side. Signs that the surface itself varies are a large spread of readings from one lab along the same face, visible chatter or tool marks in one area, or a coating or handling mark that one lab traced across and the other did not.
What causes two labs to read different Ra on the same surface?
The usual causes are a different cutoff or filter, a trace taken along the lay instead of across it, a different location on the surface, and a different evaluation length or acceptance rule. Stylus tip, skid pickups, uncorrected form on curved surfaces, coatings and instrument calibration explain most of the rest.
Cutoff and filter. The roughness cutoff (λc) separates roughness from waviness. A longer cutoff lets longer wavelengths into the roughness profile, so on a surface with waviness, chatter or feed-related undulation it raises Ra; a shorter cutoff removes them and lowers it. NIST’s roughness calibrations, for example, use a phase-correct Gaussian filter as described in ASME B46.1, a 0.8 mm cutoff, a 2.5 µm short-wavelength cutoff and typically a 4 mm evaluation length (NIST, surface roughness calibrations). The ISO 4288 condition table gives the same 0.8 mm cutoff and 4 mm evaluation length for a non-periodic profile with Ra above 0.1 µm and up to 2 µm (Mitutoyo, quick guide to surface roughness measurement); for a periodic profile such as a turned or milled face, the same table picks the cutoff from the feed-mark spacing RSm instead, with 0.8 mm for RSm above 0.13 mm and up to 0.4 mm. Either way, per ISO 4288 a 0.8 versus 1.4 µm reading does not by itself push the two labs into different default cutoffs, so a cutoff difference on the reports usually means one lab chose its own. If one lab used a longer cutoff, or an older 2RC analog filter instead of the Gaussian filter, the two numbers are not measuring the same thing.
Direction relative to the lay. In the ISO profile method the profile is taken in the direction that gives the highest roughness value, which is normally at right angles to the lay of the machining marks (Mitutoyo). A trace along the feed marks of a turned face can read far lower than a trace across them, which is a classic way for a supplier report to look better than the customer’s reading.
Evaluation length and acceptance rule. Under the older ISO 4288 practice the default was the 16% rule, which lets a limited share of readings exceed the limit; the ISO 21920-1 indication rules make the max rule the default, so one reading over the limit rejects the surface. Per ISO 4288 Annex A, the 16% rule follows a sequence: a first reading below 70% of the limit passes; otherwise two more readings are taken and all three must be below the limit; failing that, nine more are taken and no more than two of the twelve may exceed it. For the max rule it calls for at least three positions where the highest values are expected, with none over the limit (Mitutoyo, summarising ISO 4288). A supplier that stopped after one reading at 60% of the limit, as the sequence per ISO 4288 allows, and a customer that rejected on one high trace under the max rule can both be following a standard, just not the same one.
Normal scatter. The same guide notes that vertical roughness parameters such as Ra and Rz vary across a surface by roughly −20% to +30% (Mitutoyo), so a single trace from each lab is not enough to call a part good or bad when the result sits near the limit.
Stylus, pickup and form. A larger stylus tip cannot reach the bottom of narrow valleys and tends to read lower. Per ISO 3274 the nominal tip radii are 2 µm, 5 µm and 10 µm, and per ISO 4288 the condition table pairs a 2 µm tip with the 0.8 mm cutoff, allowing 5 µm when Ra is above 0.5 µm, so two compliant labs can still use different tips on the same face. A skid pickup rides on the surface and acts as a mechanical filter, which can suppress or distort waviness that a skidless pickup with a software filter would include. On a bore, a shaft or a radius, a trace that is not form-corrected adds the curvature to the profile and inflates the result.
| Cause | How to spot it on the two reports | Usual direction of the error |
|---|---|---|
| Different cutoff λc | Cutoff values differ (for example 0.8 mm vs 2.5 mm) | Longer cutoff usually reads higher on a wavy surface |
| Different filter type | One report shows Gaussian, the other 2RC or an unstated filter | Either way; results not comparable |
| Trace along the lay | Trace direction parallel to feed marks on one side | Along the lay reads lower |
| Different location | Traces on different zones, or one crosses an edge, hole or mark | Either way; often a spread of readings |
| Evaluation length or rule | One trace vs several; 16% rule vs max rule | Single short trace scatters more |
| Stylus tip radius | Tip radius differs (for example 2 µm vs 5 µm) | Larger tip usually reads lower on fine surfaces |
| Skid vs skidless pickup | Pickup type differs | Skid can suppress waviness |
| Form not removed | Curved surface, no form fit stated | Uncorrected curvature reads higher |
| Part condition | Coating, oil, residue or handling damage after the supplier measured | Either way |
| Calibration | No recent check on a roughness reference specimen | Either way |
How do you diagnose which setting caused the difference?
Remeasure one marked spot, in one marked direction, with both instruments set to the same cutoff, filter, evaluation length and pickup, and check both instruments on a roughness reference specimen first. If the numbers then agree, the dispute was the method; if they still differ, it is the instruments or the surface.
Work through the check in this order, because each step removes one variable:
- Verify both instruments. Measure a calibrated roughness specimen of the kind described in ISO 5436-1 on each instrument. An instrument that misreads the specimen settles the question before any part is touched.
- Compare the settings line by line. Cutoff, short-wavelength filter, filter type, evaluation length, number of traces, tip radius, pickup type and form removal. Any difference here is enough to explain the gap.
- Mark the trace. Put a witness mark at the start point and an arrow for the trace direction on one sample part, and have both labs measure exactly there.
- Measure across and along the lay. If the across-lay reading matches the higher number, direction was the cause.
- Apply the same acceptance sequence. If the rule is the 16% rule, both labs follow the sequence per ISO 4288 of one, three, then twelve readings; if it is the max rule, both take at least three traces where the highest values are expected. Compare the full sets, not one number each.
- Map the surface. Take several traces across the face. A wide spread means the surface varies, and the drawing needs to say where, or how many traces, the requirement applies to.
- Check the part history. Confirm whether the supplier measured before or after plating, anodizing, blasting, washing or packing, and whether the customer measured the same state.
Keep the measurement uncertainty in view. The NIST calibration service builds an uncertainty budget for Ra from calibration, stylus radius, filtering, noise and the variation of the surface itself (NIST), and a production gauge in a receiving lab carries more uncertainty than a national calibration laboratory. A reading that sits just over the limit on one instrument and just under on the other may simply be within the combined uncertainty, and the drawing or the purchase agreement should say how that case is decided.
How do you resolve a roughness dispute once a lot is on hold?
Agree the method first, then remeasure, then disposition the lot against the drawing as written. Change the drawing only if it was silent or ambiguous, and change it for future lots, not by argument over this one.
The decision table below covers the outcomes a buyer usually meets. The aim is to close the lot on facts both sides accept and to make sure the same dispute cannot happen on the next order.
| What the remeasurement shows | Likely root cause | Action on this lot | Action for future lots |
|---|---|---|---|
| Same settings, readings agree and pass | Method mismatch | Release the lot | Put the settings on the drawing |
| Same settings, readings agree and fail | Process: tool wear, feed, chatter, vibration | Rework or reject per the PO | Corrective action on the process; add an in-process check |
| Same settings, readings still differ | Instrument, calibration or uncertainty | Third-party measurement on marked spots | Correlate gauges on a shared reference part |
| Readings vary across the face | Local defects or uneven finish | Measure at the functional zone | Define the zone and number of traces |
| Drawing gives only “Ra” and a value | Ambiguous callout | Measure at the standard default and record it | Revise the callout with method and rule |
| Coating applied after the supplier measured | Measured in a different state | Measure the state the drawing controls | State before or after coating on the drawing |
What should the drawing specify so roughness is measured the same way everywhere?
The drawing should name the parameter and limit, the acceptance rule, the cutoff or setting class, the evaluation length if it is not the default, the lay or trace direction, the surface or zone it applies to and whether it applies before or after any coating. It should also name the standard, because the defaults differ between the older ISO 1302 and ISO 4288 practice and ISO 21920, and ASME drawings follow their own standards.
- Parameter and limit. Write the parameter and an upper limit, and add a lower limit only where function needs one; the Ra vs Rz guide covers which parameter to choose.
- Acceptance rule. Write the rule into the callout per ISO 21920-1, as “max” or “16%”. Per ISO 21920-1 the max rule is the default, where ISO 1302 and ISO 4288 used the 16% rule, so a bare value means different things depending on which edition the reader assumes.
- Filter and cutoff. State the cutoff, or a setting class, rather than leaving the lab to pick one from the expected value. For example, a complete requirement written per ISO 21920-1 states an Ra upper limit of 0.8 µm, the max rule, a 0.8 mm cutoff and the default evaluation length.
- Direction. Add the lay symbol where the lay matters, and a note such as “measure perpendicular to lay” where a reviewer might trace the other way.
- Location. Apply the symbol to the functional surface only, and use a zone or a note for a seal land or bearing track rather than a general note covering the whole part.
- Coating state. Say whether the value applies before or after plating or anodizing; the coating guide explains why the answer changes the number.
- Standard. Name ISO 21920-1 or ASME Y14.36 and ASME B46.1 in the title block or general notes.
On the RFQ and purchase order, add what the drawing cannot carry: which surfaces need a roughness reading on the inspection report, how many parts per lot, how many traces per surface, and that the report must show the instrument settings and not only the result. For how the symbol itself is drawn, see the surface finish symbol guide.
Who checks surface roughness on a Saola order?
Roughness on machined surfaces is a matter for the lead CNC plant that turns or mills the part, while roughness after plating, anodizing or other finishes depends on the partner plant we manage for that finish. The measurement method, and who makes the reading, is agreed at the quote stage so it does not have to be argued after shipment.
Machined surfaces come off CNC turning and CNC milling at our lead CNC plant, T.K Viet Nam Precision Company Limited. Finishes are applied through the surface finishing partner plants we manage, which are not necessarily in Vietnam; the quote states where each operation runs and the country of origin. Every drawing is reviewed callout by callout, and a roughness symbol that does not state the cutoff, the rule, the direction or the coating state is raised as a question on the marked-up drawing rather than assumed. If your receiving inspection uses a particular cutoff, pickup or rule, put it on the drawing or the RFQ, and say which surfaces must appear on the first-article report with their instrument settings. Send the drawing through the quote form.
Frequently asked
Which reading is right when supplier and customer disagree?
The one taken with the method the drawing specifies. If the drawing is silent, the defaults of the standard it names apply, and both sides should remeasure to those defaults on a marked spot before anyone dispositions the lot.
Can a portable roughness tester be used for acceptance?
Yes, if it can be set to the specified cutoff, filter and evaluation length, it is checked on a reference specimen, and its pickup type suits the surface. Some portable units use a skid pickup, so state on the drawing or PO if a skidless measurement is required.
Does ISO 21920 make older roughness callouts invalid?
No. Drawings made to ISO 1302 and ISO 4288 still carry their meaning, but the default acceptance rule and some evaluation rules differ, so name the standard on the drawing so the lab knows which defaults to use.
Why does a turned face read better along the feed marks?
Along the lay, the stylus travels in the trough or on the crest of the feed marks and sees little height change. Across the lay it crosses every peak and valley, which is why the ISO profile method takes the direction that gives the highest value.
Should roughness be measured on every part?
Usually not. Measure it at first article and then by a sampling plan on the surfaces that matter for sealing, bearing or appearance, and write that plan into the PO so the supplier and receiving inspection follow the same one.
Sources
- ISO 21920-1:2021, Surface texture: Profile — Part 1: Indication of surface texture, ISO, 2021
- ISO 21920-2:2021, Surface texture: Profile — Part 2: Terms, definitions and surface texture parameters, ISO, 2021
- ISO 21920-3:2021, Surface texture: Profile — Part 3: Specification operators, ISO, 2021
- ISO 4288:1996, Rules and procedures for the assessment of surface texture (withdrawn, replaced by ISO 21920-3), ISO, 1996
- ASME B46.1-2019, Surface Texture (Surface Roughness, Waviness, and Lay), ASME, 2019
- Quick Guide to Surface Roughness Measurement, Bulletin No. 2229 (summary of ISO 4287, ISO 4288, ISO 3274 and ISO 1302 conditions), Mitutoyo, accessed October 2026
- NIST Surface Roughness and Step Height Calibrations: Measurement Conditions and Sources of Uncertainty (Vorburger et al.), NIST, accessed October 2026