What Surface Finish (Ra)
Do You Actually Need?
Ra is one number describing an entire surface. It's the most useful measurement in finishing and the most misunderstood — because an average, by definition, throws away everything that made the surface interesting. Here's what Ra tells you, what it hides, and how to stop paying for finish nobody asked for.
Two questions get asked about surface finish. Only one of them is the right one.
The wrong question is "how smooth can we get it?" The right one is "what does this surface have to do?" — because a surface that seals, a surface that slides, a surface that holds paint, and a surface that just has to look decent all want different numbers, and two of them get worse as you polish them.
Ra is the number the drawing uses to answer that. It's not the only one, it's not always the right one, and it's routinely specified tighter than the part needs by people who assume smoother is better. This guide covers what it actually measures, where it lies to you, and what number the job actually calls for.
What Ra Actually Measures
Drag a stylus across a surface and it traces a wandering line — peaks and valleys around some mean height. Ra is the arithmetic average of how far that line deviates from the mean, ignoring whether the deviation was up or down.
Ra (roughness average) is the arithmetic mean of the absolute vertical deviations of the surface profile from its mean line, measured over a defined evaluation length. Reported in microinches (µin) in the US and micrometers (µm) elsewhere.
Three things follow from that definition, and each one matters:
- It's an average. Every peak and valley in the trace gets folded into one number. Nothing survives except the mean deviation.
- It ignores sign. A peak sticking up 40 µin and a valley cutting down 40 µin contribute identically. Ra cannot distinguish a surface covered in sharp peaks from one covered in deep pores.
- It ignores direction. Ra says nothing about which way the scratches run — that's lay, and it's a separate spec that matters enormously on sealing and sliding surfaces.
None of that makes Ra a bad measurement. It makes it a summary — and you have to know what a summary leaves out before you can trust it.
Three Surfaces. Identical Ra.
These aren't illustrations. They're three profiles with the same Ra to six decimal places — the arithmetic is exact. Put them on a print and all three pass the same callout.
Three profiles, one Ra value — equal to six decimal places, not approximately. The averaging is what makes them equal: the gouged surface's three deep valleys are diluted across the whole trace, and the spiked surface's peaks are diluted the same way.
Now put them in service
- Uniform — the good one. Consistent texture, predictable behavior, holds a lubricant film, seals against a gasket. This is what "Ra 32" is supposed to mean.
- Gouged — a leak path. Three deep valleys crossing a sealing face is a part that fails a pressure test no matter what the Ra callout says. On a fatigue-critical part, each valley is a stress riser and a crack initiation site.
- Spiked — a galling problem. Sharp peaks on a sliding surface carry the entire load on a fraction of the contact area. They plastically deform, weld micro-junctions, and tear. The part passes inspection and eats itself in service.
The Takeaway, Said Plainly
Ra tells you how rough a surface is on average. It does not tell you whether the surface is any good. An Ra callout with no lay requirement, no Rz limit, and nobody looking at the part is a spec that can be met by a surface that fails in the field. That's not a reason to abandon Ra — it's a reason to know what you're asking for when you write it.
Ra, Rz, Rq, Rt — And Why They Don't Convert
The figure above is exactly why other parameters exist. Each one catches something Ra throws away.
| Parameter | What It Measures | What It Catches That Ra Misses |
|---|---|---|
| Ra | Arithmetic average deviation from the mean line | Nothing extra — it's the baseline. Stable, repeatable, insensitive to isolated features. That insensitivity is both the feature and the flaw. |
| Rz | Average of the maximum peak-to-valley heights across sampling lengths | Extremes. Rz responds to the deep gouge and the tall spike that Ra averages away. If your failure mode is a single defect, Rz is the parameter that sees it. |
| Rq (RMS) | Root mean square of the deviations | Weights large deviations more heavily than Ra does, because it squares them. Common in optics. Typically runs modestly higher than Ra on a normal surface. |
| Rt / Rmax | Total or maximum peak-to-valley height | The single worst feature in the trace. Unforgiving and sensitive to one-off damage — which is the point when one-off damage is the thing that kills the part. |
↔ Scroll table horizontally on mobile
There Is No Ra-to-Rz Conversion
You'll see rules of thumb — Rz runs somewhere around four to seven times Ra on typical machined surfaces. That's an observation about common surfaces, not a conversion factor. The ratio depends entirely on the character of the profile, which is exactly the information Ra discarded. Look at the three traces above: same Ra, three different Rz values. Any chart that promises to convert between them is telling you something it cannot know.
If a print calls out Rz, measure Rz. Converting from an Ra reading isn't a shortcut — it's a guess.
The Numbers: µin, µm, and N-Grades
Three notations for the same thing. Microinches in US shops, micrometers everywhere else, and ISO grade numbers on drawings that use them. The standard series doubles at each step.
| Ra (µin) | Ra (µm) | ISO Grade | Character |
|---|---|---|---|
| 2000 | 50 | N12 | Very rough — as-cut, as-cast |
| 1000 | 25 | N11 | Rough |
| 500 | 12.5 | N10 | Rough machining |
| 250 | 6.3 | N9 | Medium machining |
| 125 | 3.2 | N8 | Standard machined finish |
| 63 | 1.6 | N7 | Good machined finish — the common default |
| 32 | 0.8 | N6 | Fine — grinding territory, typical satin |
| 16 | 0.4 | N5 | Very fine |
| 8 | 0.2 | N4 | Precision ground / honed |
| 4 | 0.1 | N3 | Lapped |
| 2 | 0.05 | N2 | Superfinished |
| 1 | 0.025 | N1 | Mirror / optical |
↔ Scroll horizontally on mobile · µin values are the standard nominal series
The exact conversion is 1 µm = 39.37 µin. The table uses the standard nominal series — 6.3 µm is 248 µin and everyone calls it 250. Don't be the person who argues about it.
Every Step Down Doubles the Fineness — and More Than Doubles the Cost
This is a geometric ladder, not a linear one. Going from 125 to 63 is one step. Going from 125 to 8 is four steps, and it means a different process, different equipment, more operations, and more inspection. The number on the print looks like a small edit. The shop floor experiences it as a different job.
What Each Process Actually Delivers
Typical achievable Ra by process. Ranges are broad because setup, tooling condition, material, and operator all move the number — treat these as where the process lives, not what it guarantees.
| Process | Typical Ra (µin) | Typical Ra (µm) | Note |
|---|---|---|---|
| Flame / plasma cutting | 500 – 2000 | 12.5 – 50 | A starting point, not a finish |
| Sand casting | 250 – 1000 | 6.3 – 25 | Varies with sand and pattern |
| Sawing | 250 – 1000 | 6.3 – 25 | |
| Rough turning / milling | 125 – 500 | 3.2 – 12.5 | Feed rate dominates |
| Drilling | 63 – 250 | 1.6 – 6.3 | |
| EDM | 32 – 250 | 0.8 – 6.3 | Settings drive it hard |
| Finish turning / milling | 32 – 125 | 0.8 – 3.2 | Where most parts land |
| Reaming | 32 – 125 | 0.8 – 3.2 | |
| Abrasive belt / non-woven finishing | 16 – 125 | 0.4 – 3.2 | Grade and pressure dependent |
| Grinding | 8 – 63 | 0.2 – 1.6 | The workhorse for fine finishes |
| Honing | 4 – 32 | 0.1 – 0.8 | Controls lay as well as Ra |
| Polishing | 1 – 16 | 0.025 – 0.4 | |
| Lapping | 1 – 8 | 0.025 – 0.2 | |
| Superfinishing | 0.5 – 8 | 0.012 – 0.2 | Bearing races, seal journals |
↔ Scroll horizontally on mobile · Typical ranges, not guarantees
Read the Overlaps
Finish milling and grinding both cover 32–63 µin. If your part only needs 63, a grinding operation is a step you're paying for that finish-turning already gave you. The cheapest way to hit an Ra is to spec a number the process you're already running can hold. Look at where the ranges overlap before you add an operation.
What Ra Does the Job Need?
Work from function, never from instinct. Here's the honest answer by what the surface has to do — including the cases where smoother makes it worse.
| Surface Function | Direction | What Actually Governs |
|---|---|---|
| Non-critical / as-machined | 125 – 250 | Nothing but appearance and handling. If it doesn't seal, slide, seat, or show, don't spec it tighter. Most surfaces on most parts belong here. |
| Cosmetic / decorative | Appearance-driven | Judged by eye, not by instrument. Uniformity and lay matter more than the number — a consistent 60 looks better than a blotchy 30. Match the adjacent panel. |
| Paint / coating adhesion | Too smooth = failure | Coatings need anchor profile to grip. Polish past what the system wants and you get adhesion failure. The coating manufacturer's TDS governs — it overrides any general guidance, including this table. |
| Static seal / gasket face | 32 – 125 | The gasket has to conform into the texture. Lay matters more than Ra — concentric or radial tool marks across a face are a leak path regardless of the number. Follow the seal manufacturer's spec. |
| Dynamic seal / O-ring journal | Has a floor and a ceiling | Too rough abrades the elastomer. Too smooth starves the interface of lubricant film and the seal wears anyway. This is a window, not a minimum. Seal manufacturer's spec, every time. |
| Bearing / sliding surface | Too smooth = galling | Surfaces need to retain oil. A mirror-finish journal has nowhere to hold a film. Honed surfaces use a cross-hatch lay specifically to trap lubricant — that's texture doing a job. |
| Fatigue-critical | Smoother is better | One of the few places the instinct is right. Valleys are stress risers and crack initiation sites. Here you care about Rz and Rt more than Ra — a single deep scratch is the failure, and Ra won't see it. |
| Hygienic / cleanable | Smoother is better | Crevices harbor product and resist cleaning. Food, dairy, pharma, and bioprocess work are governed by specific standards — 3-A, ASME BPE and their equivalents set the requirement, and the current edition governs. Don't work from a blog table. |
| Press / interference fit | 63 – 125 | Peaks shear off during assembly, which changes the actual interference from what you calculated. Rougher than planned means less grip than the math said. |
| Weld prep | Clean > smooth | Contamination is the enemy, not roughness. A clean 125 welds better than a greasy 32. |
↔ Scroll horizontally on mobile · Directional guidance — the applicable spec always governs
Four of Ten Get Worse When You Polish Them
Coating adhesion, dynamic seals, bearing surfaces, and cosmetic uniformity all fail or degrade if you chase a smaller number. That's not a technicality — it's most of the surfaces that actually do something. Smoother is not better. Smoother is just smoother, and it always costs more.
Lay: The Spec Everyone Forgets
Lay is the direction of the dominant surface pattern — which way the scratches run. Ra is blind to it, and on sealing and sliding surfaces it can matter more than the number.
A sealing face turned with concentric tool marks has a continuous spiral groove running around it. That's a leak path, and it will leak at Ra 32 while a face with random lay seals fine at 63. The number was never the problem.
| Symbol | Lay Direction | Typical Of |
|---|---|---|
| = | Parallel to the plane of projection | Turning, shaping, planing |
| ⊥ | Perpendicular to the plane of projection | Turning, shaping |
| X | Angular in both directions — cross-hatch | Honing. Deliberately traps lubricant. |
| M | Multidirectional | Lapping, blasting, non-woven finishing |
| C | Approximately circular to the center | Facing operations |
| R | Approximately radial from the center | |
| P | Particulate, non-directional, protuberant | Shot peening, some coatings |
If Lay Matters, Say So on the Print
An Ra callout with no lay symbol tells the shop that any direction is acceptable — and they'll produce whatever direction the process naturally gives. If the function depends on lay, the drawing has to carry it. "They should have known" isn't a specification.
Six Ways Ra Costs You Money
- Speccing tight "to be safe." The most expensive habit in this article. Every step down the ladder can add an operation, equipment, and inspection time. If you can't name the function that needs it, you don't need it.
- Assuming smoother is better. Coatings, dynamic seals, and bearing surfaces all get worse. You're not adding margin — you're introducing a failure mode and paying for the privilege.
- Converting Ra to Rz off a chart. The ratio depends on profile character, which Ra threw away. If the print says Rz, measure Rz.
- Specifying Ra when the failure mode is a single defect. Ra averages one deep gouge into nothing. If a scratch kills the part, Rz or Rt is the parameter — Ra will pass it every time.
- Leaving lay off the print. Then discovering the sealing face was turned with concentric marks and leaks at spec.
- Ignoring the spec that actually governs. Coating TDS, seal manufacturer data, and hygienic standards all override general guidance. That includes this article.
Answers, Short Version
What does Ra 32 mean?
An average roughness of 32 microinches (0.8 µm) — the arithmetic mean deviation of the surface profile from its mean line. It's a fine finish, roughly what grinding or good non-woven finishing produces, and typical of a satin surface. It says nothing about lay, or about whether the surface has a deep scratch in it.
Is a lower Ra always better?
No, and this is the most expensive misconception in surface finishing. Coating adhesion needs anchor profile. Dynamic seals and bearing surfaces need texture to retain a lubricant film. Polish those past their window and you cause the failure you were trying to prevent. Lower Ra always costs more; it only sometimes helps.
How do I convert Ra to Rz?
You don't. Rz commonly runs somewhere around four to seven times Ra on typical machined surfaces, but that's an observation, not a conversion — the ratio depends on the profile character that Ra already discarded. Two surfaces with identical Ra can have very different Rz. If the print calls out Rz, measure Rz.
Ra vs RMS — what's the difference?
Ra is the arithmetic average of the deviations. Rq (RMS) is the root mean square, which squares the deviations and so weights the large ones more heavily. RMS reads modestly higher than Ra on a typical surface. RMS is common in optics; Ra dominates general engineering.
What Ra do I need before painting?
Whatever the coating manufacturer's technical data sheet says — and it's a window, not a maximum. Coatings need surface profile to grip; too smooth is a real adhesion failure mode. Never polish a substrate finer than the coating system asks for.
Can two surfaces have the same Ra and behave differently?
Yes, dramatically. A uniform texture, a smooth surface with a few deep gouges, and a flat floor covered in sharp peaks can all share one Ra value. The first seals and slides; the second leaks and cracks; the third galls. Ra averages the difference away — that's why Rz, Rt, and lay exist.
What Ra is a #4 brushed stainless finish?
Careful here — #4 is defined by appearance and process, not by a single Ra number, and different specifications attach different roughness requirements to it. It's typically in the fine range associated with medium to fine abrasive finishing, but if you need a number you need the spec you're being held to, not a general rule. Match the adjacent surface and confirm against the governing standard.
What grit gets me a specific Ra?
There's no fixed mapping. The same grit produces different Ra depending on pressure, speed, material, abrasive condition, and whether it's coated or non-woven. Grit and grade charts get you to the right neighborhood; a profilometer and a test piece get you to the number. Always prove it on scrap before you commit the part.
Built for the Work.
Priced for the Worker.
Pro-Graad is an independent brand making coated and non-woven abrasives for shops that measure performance in parts finished, not in logos on the box. Full grit and grade ranges, honest specs, no premium tax. Available direct at pro-graad.com.
Shop the Full Range
Chasing a specific finish and not sure how to get there? Tell us the material and the target.
info@pro-graad.com · pro-graad.com
This guide is general technical reference, not a substitute for your print, your customer's specification, a coating technical data sheet, or the governing standard for your industry. Where a specification applies, it governs.
