A drawing that says Ra 1.6 and a drawing that says Rz 6.3 are not describing the same surface, are not inspected the same way, and do not cost the same to produce. Ra and Rz are extracted from the same measured profile, but they answer different questions: Ra asks how rough the surface is on average, Rz asks how deep the worst features go. This guide explains both parameters as they are actually defined in the current ISO standards, shows why the definition of Rz repeated across most machining blogs has been obsolete since 1997, and gives KUSLA Prototype’s working rules for specifying roughness in a way a supplier can verify.
What a roughness number actually measures
A stylus profilometer or optical sensor traverses the surface and records a raw primary profile: height versus distance. That raw trace contains three superimposed things — form (the intended geometry and any bow or taper), waviness (long-wavelength undulation from machine vibration, spindle runout, or fixture deflection), and roughness (the short-wavelength texture left by the cutting edge or abrasive). Every roughness parameter is calculated only after form has been removed and the profile has been passed through band-pass filters that separate roughness from waviness.
That filtering step is not a footnote. It is the reason two labs can measure the same part and report Ra values that differ by 30 percent while both being correct under their own settings. The parameter symbol tells you which arithmetic was applied; the filter settings tell you which portion of the surface the arithmetic was applied to. A complete specification requires both.
Roughness matters because it governs behavior at interfaces: how a seal beds in, how a bearing surface retains lubricant, how a coating adheres, where fatigue cracks initiate, and how a press-fit behaves under load. KUSLA Prototype treats a roughness callout as a functional requirement to be interrogated, not a cosmetic preference — because on CNC machined parts the difference between Ra 1.6 µm and Ra 0.4 µm is typically a secondary operation and a measurable jump in unit cost.
Ra, defined properly — and what it cannot tell you
Ra is the arithmetic mean of the absolute deviations of the filtered roughness profile from its mean line, over the evaluation length. Expressed as an integral: Ra = (1/l) ∫ |Z(x)| dx. In plain terms, take the height of the profile above or below the mean line at every point, ignore the sign, and average it. ✅ Verified This definition is consistent across ISO 21920-2:2021, the superseded ISO 4287:1997, and ASME B46.1.
Ra dominates engineering drawings for two reasons that have nothing to do with it being the best descriptor. First, it is statistically stable: because it averages thousands of points, a single scratch or a speck of debris barely moves the number, so repeat measurements agree closely. Second, it was cheap to compute on analogue instruments decades before digital filtering existed, so it became the default and stayed there.
The same averaging that makes Ra repeatable makes it blind. A surface with a uniform, shallow, evenly spaced tool pattern and a surface carrying one deep gouge can return identical Ra values, because one deep valley contributes almost nothing to a mean taken over the full evaluation length. If the failure mode you care about is a leak path under a gasket, a stress riser, or a plating void, Ra alone will not protect you.
The Rz definition most machining guides still get wrong
Search “Rz” and the overwhelming majority of supplier blogs will tell you Rz is the “ten-point height of irregularities” — the average of the five highest peaks and the five lowest valleys. ❌ Error That definition was withdrawn from ISO practice in 1997.
Under ISO 4287:1997 and carried into ISO 21920-2:2021, Rz is the maximum height of the profile: within one sampling length, take the single largest peak height (Rp) and the single largest valley depth (Rv), and sum them. Rz = Rp + Rv. When five sampling lengths make up the evaluation length, the reported Rz is the mean of the five individual Rz values. ✅ Verified
The obsolete ten-point parameter still exists, but it is now named Rzjis (from the Japanese standard) or referred to as Rz DIN from the old DIN 4768 usage, and it must be labelled as such. On the same profile, the two definitions do not agree — the ISO maximum-height Rz is systematically the larger of the two, because it is driven by single extremes rather than by a five-point average that dilutes them.
ISO 21920-2:2021 has replaced ISO 4287 — what changed
ISO 21920 (published December 2021, in three parts) consolidated and replaced the long-standing profile texture standards ISO 4287, ISO 4288, ISO 3274, ISO 12179, ISO 13565 and ISO 1302. ✅ Verified Most drawings in circulation today still reference the older standards, and most shop-floor instruments still default to them, so both systems will coexist for years. Three changes matter for anyone reading or writing a callout:
- Order of operations changed. Under ISO 4287 the primary profile was obtained by removing nominal form and then applying the short-wavelength λs filter. ISO 21920-2 reverses these two steps to match how modern software actually processes data. ✅ Verified
- Averaging changed for most parameters. ISO 4287 computed a parameter on each of the (default five) sampling lengths and reported the mean. ISO 21920-2 computes most parameters once, over the whole evaluation length. Peak-and-valley parameters — Rp, Rv, Rz — retain the per-section calculation because averaging over sections is what keeps them stable. ✅ Verified
- Results can shift. Because the arithmetic path differs, the same trace can yield slightly different values under the two standards. Rq is largely unaffected; shape parameters such as Rsk move the most. ⚠️ Caution — magnitude of shift is surface-dependent and should be established empirically on your own parts, not assumed from a published factor.
The takeaway is not that older drawings are wrong. It is that “Ra 0.8” is an incomplete instruction unless the drawing also states the standard it is written against — and increasingly, prints will need to say so explicitly.
Why a Ra value without a cutoff is not verifiable
The cutoff wavelength λc is the filter boundary that decides what counts as roughness and what is dismissed as waviness. Choose a longer cutoff and more long-wavelength undulation is admitted into the roughness profile, raising the reported Ra. Choose a shorter one and the same surface reports smoother. The number on the report is meaningless without it.
ISO 4288 provides the standard selection table for non-periodic profiles (ground, milled, turned, EDM surfaces), reproduced below. In practice, an instrument that reports 0.8 mm as its default is following this table for the mid-range of ordinary machined work.
| Ra (µm) | Rz (µm) | Cutoff λc (mm) | Sampling length (mm) | Evaluation length (mm) |
|---|---|---|---|---|
| 0.006 – 0.02 | 0.025 – 0.1 | 0.08 | 0.08 | 0.4 |
| >0.02 – 0.1 | >0.1 – 0.5 | 0.25 | 0.25 | 1.25 |
| >0.1 – 2 | >0.5 – 10 | 0.8 | 0.8 | 4.0 |
| >2 – 10 | >10 – 50 | 2.5 | 2.5 | 12.5 |
| >10 – 80 | >50 – 200 | 8.0 | 8.0 | 40.0 |
✅ Verified Values per ISO 4288 cutoff selection for non-periodic profiles; the evaluation length is five sampling lengths by default.
Two operational rules follow. First, the table is circular by design — you need an approximate Ra to pick the cutoff that measures Ra — so the correct procedure is to take a trial measurement, land in a band, and re-measure at that band’s cutoff. Second, the traverse must be long enough: an evaluation length of 4 mm plus pre- and post-travel means a 0.8 mm cutoff measurement needs roughly 5 mm of accessible flat surface. On small features — a 3 mm boss face, a narrow groove floor — that surface does not exist, and the honest answer is to specify a shorter cutoff and accept a different number, not to pretend the standard traverse was possible.
The Rz/Ra ratio: a process fingerprint, not a conversion
On ordinary machined surfaces the ratio Rz/Ra typically falls between roughly 4 and 7. ⚠️ Caution — this is an observed empirical band across common processes, not a standardized constant, and no ISO document sanctions converting between the two. The value of the ratio is diagnostic rather than arithmetic. Read it as a signature of what the process is doing:
| Observed Rz/Ra | What the surface is telling you | Typical cause |
|---|---|---|
| ~4 or below | Highly uniform, periodic texture; peaks and valleys are consistent | Fine turning or single-point boring with a stable feed and a sharp insert |
| ~4 – 7 | Normal, well-behaved machined or ground surface | Routine milling, turning, grinding within control |
| ~8 – 12 | Isolated deep features hiding under an acceptable average | Built-up edge, a chipped insert, chip re-cutting, or a torn grain in the workpiece |
| Above ~12 | The surface is defined by defects, not by the cutting pattern | Tool breakout at an exit edge, embedded abrasive, porosity or inclusions opening at the surface |
This is why KUSLA Prototype records both parameters during in-process checks on sealing and bearing surfaces even when the customer print calls out only Ra. A stable Ra with a climbing Rz across a production run is the earliest visible warning that a tool is failing — the mean has not moved yet, but the extremes have. Acting on that signal before the Ra drifts out of tolerance is the difference between changing an insert and scrapping a batch.
Conversion charts and exactly where they break
Ra-to-Rz conversion tables circulate widely and are genuinely useful for one narrow purpose: estimating what an unfamiliar callout roughly implies so you can sanity-check a quote or select a process. They are not valid for inspection, and a supplier who measures Ra and converts to demonstrate conformance to an Rz callout has not demonstrated conformance.
The reason is structural. Ra is a mean and Rz is an extreme; a mean cannot predict an extreme without knowing the shape of the height distribution, and the shape is exactly what neither parameter reports. Two surfaces with identical Ra and radically different Rz are not an edge case — they are the ordinary outcome of comparing a cleanly cut surface to one with intermittent damage, which is the comparison that matters most.
The related parameters are worth knowing for the same reason. Rq (root-mean-square roughness) runs about 1.1 × Ra on typical machined surfaces ⚠️ Caution — empirical approximation and is preferred in optics and semiconductor work because squaring the deviations weights outliers more heavily and it feeds directly into scattering models. Rt is the total height of the profile across the entire evaluation length — a single worst peak-to-valley figure, with no averaging at all, making it the strictest and least repeatable of the common parameters. Rmax (Rz1max under current nomenclature) is the largest of the individual sampling-length Rz values rather than their mean, and is the right choice when a single defect anywhere on the surface is disqualifying, as it is for a dynamic sealing face.
Achievable Ra and Rz by manufacturing process
The table below gives the ranges KUSLA Prototype quotes against for prototype and low-volume work. Rz figures are shown as the range that follows from the Rz/Ra band above for a well-controlled process; they are planning figures, not guarantees.
| Process | Typical Ra (µm) | Indicative Rz (µm) | Notes on cost and control |
|---|---|---|---|
| Lapping / polishing | < 0.1 | < 0.7 | Dedicated secondary operation; geometry-limited, flat and cylindrical features only |
| Grinding | 0.1 – 0.8 | 0.5 – 5 | Standard route below Ra 0.8 on hardened steel; requires grind stock in the design |
| Turning | 0.4 – 3.2 | 2 – 20 | Ra scales with feed squared over nose radius; finish passes cost cycle time |
| Milling | 0.8 – 6.3 | 4 – 40 | Ra 1.6 is the practical economic floor for a general finish pass |
| Reaming | 0.4 – 1.6 | 2 – 10 | Bore finish and size in one operation; see KUSLA’s guide to reaming |
| Diamond turning | < 0.05 | < 0.4 | Optical-grade finish on non-ferrous alloys; see diamond turning |
| Wire / sinker EDM | 0.4 – 3.2 | 3 – 25 | Stochastic, non-directional texture; Rz/Ra ratio runs higher than for cut surfaces |
✅ Verified Ra ranges cross-checked against published process capability data and Machinery’s Handbook surface finish ranges. ⚠️ Caution Rz columns are derived planning estimates.
The choice of process drives the achievable range far more than any finishing trick applied afterwards, which is why roughness targets belong in the DFM conversation at quoting stage. The distinction between what a milling versus a turning operation can hold on the same feature is often the deciding factor in whether a callout is economical or not, and on formed parts the relevant limits come from the forming process rather than from cutting at all — a point covered in KUSLA’s sheet metal fabrication guidance.
Decoding the surface texture symbol on a drawing
The graphical symbol is a check-mark-like tick with an extended horizontal arm, and each position around it carries specific information. Reading it in order:
- Above the arm, left: the parameter and its limit, e.g. Ra 1.6. A value alone with no parameter symbol is, by long convention, Ra in micrometres — but writing it out removes all doubt.
- Prefix letters U and L: upper and lower limits. A lower limit is not redundant — some surfaces must not be too smooth, because adhesive bonds and paint keys need texture to grip.
- The filter band, written as λs–λc: e.g. 0.0025-0.8. Omitted on most prints, which is precisely the ambiguity discussed above.
- Below the arm: the manufacturing method or treatment, e.g. ground or anodized. Note that a roughness value specified after coating is a different requirement from the same value before coating, and anodizing in particular alters the measured surface.
- The lay symbol (=, ⊥, X, M, C, R) indicating the direction of the dominant pattern. Lay matters wherever motion, flow, or sealing is directional, and it is free to specify.
- A circle in the vee means material removal is prohibited; a bar across the vee means material removal is required.
One measurement rule follows from the lay symbol: the stylus must traverse perpendicular to the lay, because that is the direction in which the texture is at its maximum. Tracing along the lay on a turned part will produce a flatteringly low Ra that no one else will be able to reproduce.
Specification checklist before you release the drawing
KUSLA Prototype applies the following checklist when reviewing incoming prints. It resolves the large majority of roughness-related quoting queries before they are raised.
- Is the parameter named? “1.6” is not a specification. “Ra 1.6 µm” is.
- Is the standard named? State ISO 21920-2 or ISO 4287 — and if Rz is specified, state whether ISO maximum-height Rz or Rzjis is intended.
- Is the cutoff stated, or is the ISO 4288 default accepted? If the default is acceptable, say so on the drawing; if the feature is too small for the default traverse, specify the shorter cutoff yourself.
- Is the requirement on the right surfaces only? A blanket note applying Ra 0.8 to every surface on a part multiplies cost across faces that never touch anything. Apply tight finishes locally.
- Does the parameter match the failure mode? Sealing, fatigue, and coating adhesion are governed by extremes — specify Rz or Rz1max. Friction, general appearance, and flow are governed by the average — Ra is appropriate.
- Is the finish specified before or after surface treatment? State which, whenever a coating, plating, or conversion treatment is called for.
- Is the lay specified where it functionally matters? If it does not matter, leave it off rather than over-constraining.
- Is the value achievable by the intended process? Cross-check against the process table above before release; upgrading a milled face to Ra 0.4 means adding a grinding operation, fixturing, and stock.
Related KUSLA resources
- Manufacturing services overview — prototyping and low-volume production capabilities
- CNC machining services — tolerance and finish capability by feature type
- Diamond turning guide — reaching sub-0.05 µm Ra on optical components
- What is reaming — bore finish and size control in one operation
- CNC milling vs turning — how the process choice constrains achievable finish
- Sheet metal fabrication — finish considerations on formed and cut parts
Frequently asked questions
Need a part held to a specific Ra or Rz?
Send your drawing and KUSLA Prototype’s engineering team will confirm the achievable finish, the process route, and the inspection method before you commit to an order.
Request a QuoteData verification legend: ✅ Verified two or more independent sources agree · ⚠️ Caution single source or empirical approximation · ❌ Error conflicts with the governing standard.