Surface Finish on CNC Parts: What Ra Actually Measures, and How to Specify It Without Overpaying

Surface finish is the most over-specified requirement in CNC procurement. A drawing calls out Ra 0.4 µm because "smooth is better," the quote comes back 40% higher than expected, and nobody is sure whether the cost is real or whether the shop is padding the number.

The problem is rarely the shop. It's that Ra is a single averaged number being asked to describe a surface that has depth, direction, and waviness — and the process that produces Ra 0.4 µm may be entirely different from the one that produces Ra 1.6 µm. Understanding which process you are actually buying is the difference between a fair price and an expensive habit.

What Ra actually measures

Ra is the arithmetic mean deviation of the roughness profile from its mean line, measured over a sampling length. Three things follow from that definition, and each one has commercial consequences.

It is an average, so it hides peaks. Two surfaces can share Ra 1.6 µm while one has occasional deep scratches and the other is uniformly textured. The first can fail a seal; the second will not. Ra alone cannot distinguish them.

It ignores direction. A turned surface and a milled surface at equivalent Ra behave differently in sliding contact because the lay runs in different directions. A seal that works on a circumferentially turned shaft can leak on a face-milled one at the same Ra value.

It is measured over a cutoff length. Change the cutoff and the number changes without the part changing. This is why two inspection reports on the same part can disagree, and it is worth asking which cutoff and which standard a supplier used before treating a discrepancy as a quality problem.

Ra also has companions that matter in specific applications:

Parameter	What it captures	Where it matters
Ra	Average roughness	General acceptance criterion
Rz	Mean peak-to-valley height	Sealing faces, fatigue-sensitive parts
Rmax	Single worst peak-to-valley	Coatings, thin-wall pressure parts
Rmr / tp	Material ratio at a given depth	Bearing surfaces, wear resistance

If a drawing specifies only Ra on a sealing or bearing surface, that is a specification gap, not a supplier error.

The process ladder: what each Ra band actually costs

The practical value of a finish specification is that it maps to a process. Below is how the bands behave on typical CNC work in aluminum and stainless steel. These are process relationships, not quotes — the multiplier depends on geometry, material, and batch size.

Ra (µm)	Typical process	Practical note
6.3 – 3.2	Roughing / semi-finishing passes	Normal for non-functional faces and internal structure
3.2 – 1.6	Standard finishing pass	The default for most functional machined surfaces
1.6 – 0.8	Finishing with reduced stepover and sharper tooling	Achievable on most 3-axis and 5-axis work with the right cutter
0.8 – 0.4	Fine finishing, light radial cuts, stable setup	Sensitive to tool wear and vibration; needs a rigid setup
0.4 – 0.2	Precision finishing or grinding	Milling becomes unreliable; grinding or lapping territory
Below 0.2	Lapping, honing, superfinishing	A separate process, often a separate supplier

The cost curve is not linear. Moving from 3.2 to 1.6 µm is usually a parameter change. Moving from 0.8 to 0.4 µm is often a tooling change plus a time increase per pass. Moving below 0.4 µm is frequently a process change — a different machine, a different department, sometimes a different vendor — and that is where the price jumps.

One practical rule: specify the coarsest finish that satisfies the function. For most non-sealing, non-sliding machined faces, 3.2 µm or 1.6 µm is entirely adequate. Requiring 0.8 µm on a bracket that only needs to look clean buys nothing and costs real money.

Where finish is genuinely functional

Not all tight finish callouts are waste. These are the cases where the extra cost is justified:

• Sealing faces. O-ring grooves and gasket surfaces need controlled Rz, because it is the peaks that create leak paths, not the average.
• Sliding and bearing surfaces. Wear rate depends on the material ratio curve, not Ra alone.
• Fatigue-critical parts. Surface irregularities act as stress risers. Rmax matters more than Ra here.
• Optical and flow surfaces. Light and fluid both respond to features smaller than the Ra average suggests.
• Coating and bonding substrates. Adhesion depends on the actual surface texture, and an over-polished surface can be worse for bonding than a moderately textured one.
• Medical and cleanroom parts. Cleanability is a function of surface topography, and Ra is a proxy rather than the whole answer.

In every one of these cases, the drawing should say why the finish is required. A note like "Ra 0.4 µm, sealing face against EPDM O-ring" tells the process engineer what to protect and where to spend effort. A bare "Ra 0.4" on a general note invites either over-processing or an argument at inspection.

The five variables that actually move the number

When a shop cannot hold a finish, the cause is almost always one of these, roughly in order of frequency:

1. Tool condition. A worn edge burnishes rather than cuts, and the finish degrades before the dimension does. Tool life management is a finish control, not just a cost control.
2. Vibration and rigidity. Chatter shows up as periodic marks in the lay. Fixing it means changing tool overhang, reducing radial engagement, or improving workholding — not slowing the spindle to a crawl.
3. Stepover. On contoured 5-axis surfaces, the visible finish is largely a stepover artifact. Finer stepover means more passes means more time, which is the honest reason fine finishes cost more.
4. Coolant and chip evacuation. Recutting chips scores the surface. On deep pockets and Swiss-type work this is often the entire problem.
5. Material behavior. Aluminum smears and builds up on the edge; stainless work-hardens; titanium generates heat at the edge. The same parameters do not transfer between materials.

Notice that only one of these is a machine capability issue. Most finish problems are setup and tooling problems, which means they are solvable without buying a new machine — a useful thing to know when a supplier says a finish is "not possible."

How to write a finish requirement that gets quoted correctly

A finish callout that produces an accurate quote has four parts:

1. The parameter and value, with standard referenced — for example Ra 1.6 µm per ISO 4287.
2. The surface it applies to, either by leader line or by a general note that scopes it. Blanket notes across an entire drawing are the most common cause of inflated quotes.
3. The measurement method and cutoff, especially where Ra and Rz could disagree.
4. The reason, in a few words. It costs nothing to write and it prevents the wrong process from being selected.

On the receiving side, a supplier who can explain which process they will use to reach a finish — and what happens if the first pass misses — is more useful than one who simply says yes to every number on the drawing.

On our own floor

At CHX Precision Metal in Shenzhen, finish requirements are handled as a process decision rather than a parameter tweak. The floor runs 220 CNC machines — 100 Taiqun T600 5-axis machining centers, 50 CNC lathes, and 70 Tsugami Swiss-type lathes — with in-house precision grinding to ±0.005 mm when a finish callout crosses below what milling can hold reliably. Dimensional tolerance is ±0.005 mm in normal production. The plant is ISO 9001:2015 and ISO 14001:2015 certified, prototypes run in 3–7 days and production in 7–15 days, with MOQ from 1 piece. If you are working through a finish specification on a sealing or sliding surface, the capability details and contact routes are on our custom CNC machining manufacturer page.

The short version: Ra is a starting point, not a specification. Name the function, choose the coarsest band that serves it, and tell the shop why — that single habit removes most of the cost and most of the arguments.