You have a drawing, a material, and a quantity. The supplier quotes a good unit price for the full volume — and asks whether you want a sample first. A sample costs extra time and a higher price per piece. Skipping it saves both. So which is right?
Order a sample when the cost of being wrong is higher than the cost of the sample. That means a new supplier, a new material, a tight tolerance, a safety-critical part, or any feature you have never had made before. Skip the sample only when the part is a repeat of a proven order, the geometry is simple, and the tolerances are loose — and even then, ask for a first-article report on the first production batch. A sample is not a formality; it is the moment you discover, cheaply, what volume production would discover expensively.
How to use this guide: the decision table tells you whether to order a sample. The acceptance checklist tells you what the sample must prove and what evidence to require with it. The last section tells you what to put in the sample order so the supplier prices the right thing.
What a sample order actually is
Three terms get mixed up, and the confusion causes real mistakes — so fix the meanings before you order anything:
- A sample (or prototype) order is a small quantity of parts machined to your released drawing, from the production material, so you can check fit, function, and finish before committing to volume. Its purpose is learning, and its quantity is small — enough pieces to test in assembly, test functionally, and keep one as the approved reference. Three to five pieces is a practical starting point for most parts: one for the test rig, one for the assembly, one kept as the approved sample.
- A first-article inspection (FAI) is not a quantity of parts. It is an inspection and a report: the first pieces of a production batch are measured in full against the drawing, and the report is sent to you before the rest of the batch is made. A sample order can include a first-article report, and a volume order should include one too — but the report and the sample are different things, and you should ask for each by name.
- A golden sample is one approved part, kept under your control (and one kept by the supplier), that becomes the physical standard the production run is judged against. It is not a souvenir. It is the reference for appearance, finish, and fit when a later batch is disputed.
A sample order is also not a 3D print. A machined sample from the actual production material, to the released drawing, tests the things a plastic print cannot: real wall thicknesses, real threads, real surface finish, and real behaviour under load. If the question you need answered is "does this geometry work in this material," a machined sample answers it. If the question is only "does this shape look right," a print may be enough — and cheaper.
Should you order a sample? Decide by risk, not by habit
Some buyers order a sample for every part, including repeats. Some never order one, including for new suppliers. Both habits waste money — the first on samples nobody needed, the second on volume orders that should never have been placed. Decide by risk:
| Situation | Sample or straight to volume? | Why |
|---|---|---|
| New supplier, first order | Sample | You are testing the supplier as much as the part: their drawing reading, their material sourcing, their inspection, their communication. A sample is the cheapest supplier test there is. |
| New material or new grade | Sample | A material you have not used behaves differently under the cutter and in service. The sample proves machinability, finish, and function in the real material — not on a data sheet. |
| Tight tolerance (below your usual range) | Sample | Tight tolerances are where cost and risk concentrate. Our CNC tolerance guide explains what tighter tolerance costs; the sample shows whether this supplier can hold it on your feature, not in general. |
| Safety-critical, sealing, or load-bearing feature | Sample | If failure has consequences, the sample is where you prove the feature — with measurement, not with trust. Ask for the inspection report with the sample. |
| Complex geometry, deep features, thin walls | Sample | Access, tool reach, deflection, and fixturing are the risks. The sample shows whether the supplier’s setup can actually reach and hold the feature. |
| Repeat order, same supplier, same drawing, proven part | Straight to volume — with a first-article report on the first batch | The process is proven. What you need is confirmation that this batch matches the last one, and a first-article report on the first pieces gives you that without a separate sample order. |
| Simple geometry, loose tolerances, non-critical part | Straight to volume — with a dimensional check | The risk is low and the sample would teach you little. Still ask what inspection is included, and get it in writing. |
| New part, but urgent and low risk | Sample, but say so | Urgency is a reason to order the sample quickly, not to skip it. Tell the supplier the sample is on the critical path and ask what can be checked in parallel. |
One rule covers the grey areas: if you cannot describe, in one sentence, what the sample would prove, you probably do not need one — but if you cannot describe what would go wrong in volume, you probably do. And if the supplier discourages a sample for a new, tight-tolerance, or safety-critical part, treat that as information about the supplier, not about the sample.
What the sample must prove
A sample that arrives with no measurements proves only that the supplier can ship a part. Before you order, decide what the sample has to prove — and write it into the order. Four categories cover most parts:
- Dimensions on the critical features. Not every dimension — the critical ones: the bores that must fit, the faces that must seal, the hole patterns that must align, the datums the assembly references. Name them in the order. Our CNC tolerance guide explains how to choose which features carry tight tolerances; the sample is where you check them.
- Material. The sample must be machined from the production material and grade — not a substitute the supplier had in stock. Ask for the material certificate with the sample. A sample in the wrong material proves geometry and nothing else.
- Surface finish on the faces that matter. A sealing face, a bearing surface, and a cosmetic face have different finish requirements, and the sample is where you check them. Our CNC surface finish guide explains Ra values and drawing callouts; ask for a roughness reading on any sealing face, not a visual judgement.
- Function and assembly. The sample must fit the mating part, assemble with the real fasteners, and do its job. A sample that measures correctly but does not assemble has still failed. Test it in the real assembly, with the real mating parts, before you approve it.
A DFM review of radii, pockets, walls and setups before the sample is ordered is what keeps the sample from proving a design that is expensive to make.
If the sample is for a part that will later be molded — a machined prototype of a part that will become an injection-molded product — be clear about what it can and cannot prove. A machined prototype proves geometry, fit, and assembly. It does not prove how the part will behave when molded: shrinkage, weld lines, and gate position are molding questions, and they are answered by a mold trial, not by a machined part.
The acceptance checklist: what evidence to require with the sample
Approve the sample against evidence, not against a look. A part that looks right and measures wrong is a wrong part. Ask for the following with the sample order, and do not approve the sample until you have them:
- A measured report against the drawing. Every critical dimension measured and recorded against its tolerance — one row per part, not one row for the whole batch. A report with a single row for five parts is a report about one part. Ask for a ballooned drawing (a drawing with numbered annotations for each critical dimension) tied to the measured values, so you can see which measurement belongs to which feature.
- The material certificate. The grade, the supplier, and the batch. For metals, the certificate should confirm chemical composition and mechanical properties; for plastics, the material data sheet for the grade supplied. A certificate of conformance that only declares compliance, without test results, is weaker evidence than a certificate with measured values — and the difference matters for critical parts.
- A surface reading on sealing and cosmetic faces. A roughness value, measured, on the faces where finish matters — not "smooth" or "as machined."
- Photos of the parts. Not as a substitute for measurement, but as a record of appearance, deburring, and any visible defects.
- The inspection scope in writing, before the order. This is the point buyers most often miss: a "standard inspection" may include no documentation at all. Xometry, the US manufacturing marketplace, states this plainly in its inspection options: a standard inspection is a simple dimensional check and no documentation will ship with your parts — to receive a dimensional inspection report correlated to your drawing, you need to upgrade to a formal, CMM, or first-article inspection, and if you want one part and stop for approval, you should select a custom inspection. Xometry UK’s inspection reports page adds the detail: its reports follow ISO 2859-1 sampling guidelines, a first-article inspection report by default includes a full measurement report plus material confirmation, and ballooned drawings are required. Those are Xometry’s published terms — your supplier’s terms will differ, which is exactly why you should ask for them in writing before you order.
One more check on the report itself: confirm the units, the datum references, and the instrument used for each measurement. A report that does not say how a dimension was measured — caliper, micrometer, or CMM — does not let you judge whether the measurement is good enough for the tolerance. Geometric tolerances (position, flatness, profile) need a CMM; a caliper cannot verify them. If your drawing carries GD&T callouts, say so in the order and ask how they will be measured.
What a sample cannot prove
A sample is powerful evidence about the part you hold. It is weak evidence about the parts you have not yet received. Three limits matter:
- A sample does not prove long-run repeatability. Five good parts prove the supplier can make five good parts. They do not prove the process will hold tolerance across five hundred, across tool changes, across material batches, across operators. Repeatability is proven by production data — measurements across consecutive batches — not by a sample. Do not ask a five-piece sample for statistical proof it cannot give, and do not accept a capability number computed from five pieces as if it described production.
- A sample does not prove the supplier’s process is in control. It proves one setup, on one day, on one machine. Ask what changes between the sample and volume — a different machine, a different fixture, a different operator, a different material batch — and what will be re-checked when it changes.
- A sample does not prove the design is right. It proves the part as drawn can be made. Whether the design survives its real loads, temperatures, and chemicals is a different question, answered by functional testing — and a sample that passes dimensional inspection can still fail in service. Test the sample the way the product will be used, not just the way the drawing is written.
None of this makes the sample useless. It makes the sample honest: use it for what it can prove — geometry, material, finish, fit, and the supplier’s ability to read your drawing — and use the first production batch, with its first-article report, to prove the rest.
Keep one approved part: the golden sample
When the sample passes, do not send all of it back, and do not put all of it into service. Keep one approved part under your control, labelled with the part number, revision, material, and approval date — and ask the supplier to keep one too. That part is the golden sample: the physical standard the production run is judged against.
The golden sample matters most when a later batch is disputed. A dimension can be argued; a finish can be argued; a part that both sides signed is much harder to argue with. Name the golden sample in the production purchase order, and state that production parts will be compared against it for appearance and fit. If the design changes, the golden sample changes with it — retire the old one, approve a new one, and record the revision. A golden sample from a superseded revision is worse than no sample at all, because it invites the supplier to make the wrong part correctly.
What to put in the sample order
A sample order that says "please make 5 pieces" will get you five pieces of something. A sample order that states the following will get you a sample you can actually approve:
- The STEP file and the 2D drawing, at the current revision, with datums, material grade, surface finish, and tolerances called out.
- The sample quantity, and what each piece is for (test, assembly, retained reference).
- The material grade and form — and that no substitute is acceptable without your written approval.
- The inspection evidence required: measured report against a ballooned drawing, material certificate, surface reading on the faces you name, and photos.
- The acceptance rule: the sample is approved by you, in writing, and volume does not start until that approval is given.
- The path to volume: the sample price, the volume price at the quantities you expect, and whether the setup, programming, and fixturing from the sample carry into the volume order.
On price: expect the sample’s unit price to be higher than the volume price. Programming, fixture preparation, machine setup, and first-article measurement are fixed costs, and on a five-piece order they are shared by five parts instead of five hundred. That is normal, and it is not a reason to skip the sample — it is the price of finding out cheaply. What matters is the total cost of the sample against the total cost of a volume order made wrong. Our CNC machining guide explains the cost drivers — setup, fixturing, tool access, material, finish, and inspection — and a sample order is where you see them itemised for your part.
As a published reference for what one supplier’s standard tolerances look like: Protolabs, the US manufacturer, states its standard CNC machining tolerance as ±0.005 in. (0.13 mm), with a standard surface roughness of 63 µin. for flat surfaces and 125 µin. for curved surfaces, and accepts GD&T callouts — see its CNC machining tolerances guide. Those are Protolabs’ published standard offering, not Cavity Mold’s capability and not a promise for your part. They are useful as a sense of what "standard" means at a large supplier — and as a reminder to ask your own supplier what their standard is, in writing, before you compare prices.
Frequently asked questions
How many sample pieces should I order?
Enough to test fit, test function, and keep one as the approved reference — three to five pieces is a practical starting point for most parts. One piece answers one question; if it goes to the test rig, you have nothing left for the assembly or the shelf. Order more if the part needs destructive testing or if several people need to evaluate it.
Can I skip the sample if the supplier is recommended by someone I trust?
A recommendation tells you the supplier is capable. It does not tell you they read your drawing correctly, sourced your material, or held your tolerance. Order the sample — and tell the supplier the recommendation is why you chose them, not why you skipped the check.
The sample passed but the first production batch failed. What went wrong?
Something changed between the sample and the batch: machine, fixture, operator, material batch, or settings. Ask for the first-article report on the batch and compare it with the sample report, feature by feature. The difference between the two reports is usually the cause — and it is why the volume order should always include a first-article report, even after an approved sample.
Should the sample be free?
Expect to pay for it. A supplier who offers a free sample has still paid the setup and programming cost, and that cost appears somewhere — in the volume price, in the inspection scope, or in the attention given to your order. A paid sample, priced transparently, is a cleaner transaction: you are buying evidence, and the supplier is selling it.
What to send, and where to send it
If you are ordering CNC machined parts for the first time — or a new part from a supplier you already use — ask for a sample quotation before the volume quotation. Through our contact page, send the STEP file and the 2D drawing with datums, the material grade, the quantities for the sample and for volume, the key tolerances and surface finish, the inspection evidence you require with the sample, your delivery region, and your timing. Tell us what the sample has to prove, and we will tell you what the sample order should include.

