ACCURACY · TOLERANCE · COST

How Hard Is 0.01 mm? The Real Link Between Machining Accuracy, Tolerance Grades and Cost

BestAI CAM Manufacturing Intelligence · published 2026-05-18 · ~11 min read · For: mechanical designers, buyers, CNC job-shop engineers

01Accuracy is not one number: the three layers—dimensional, geometric, surface

Machining accuracy is often misread as "a single tolerance number," but in manufacturing engineering it is really three layers that are independent yet interact with one another. Keeping them separate is what stops you from applying the wrong requirement and dimensioning the drawing incorrectly.

• Dimensional accuracy: the deviation of sizes such as hole diameter, length and depth from the nominal value—the familiar "±so many mm."
• Geometric accuracy (geometric tolerance): flatness, parallelism, perpendicularity, roundness, position, and so on. A hole's diameter may be fully within tolerance, yet its position relative to the datum face may still be off—something dimensional tolerance cannot control but that directly affects assembly.
• Surface accuracy (surface roughness): the microscopic undulation of the machined face, usually expressed as Ra. Sealing faces, sliding faces and cosmetic parts are especially sensitive to it.

Classic manufacturing engineering textbooks stress that a part's functional quality depends on the combination of all three, not on a single dimensional number (Kalpakjian & Schmid, 2020)^[1]. So before discussing "how accurate," you must first ask clearly: does the dimension need to be accurate, does the position need to be accurate, or does the surface need to be good? The cost structure of the three is not the same. For the logic behind geometric tolerance and GD&T callouts, see Tolerance and geometric tolerance (GD&T); for reading surface roughness, see the surface-roughness Ra guide.

02Tolerance grades and IT grades: standardizing "how accurate"

If every drawing used an absolute value like "±0.03" for tolerance, large and small workpieces would be handled inconsistently—the same ±0.03 mm is a demanding requirement on a 5 mm pin but quite loose on a 500 mm plate. To let accuracy requirements be expressed consistently, the international dimensional-tolerance system (ISO 286) introduced standard tolerance grades, commonly known as IT grades.

The IT-grade concept rests on two points:

1. The smaller the grade number, the narrower the tolerance band. IT01, IT0, IT1… as the number rises the requirement gets looser. Precision gauges sit at the very small IT grades, general machined parts mostly fall in the middle grades, and rough blanks and non-mating faces sit at the larger grades.
2. The actual tolerance value for a given grade grows as the size increases. This echoes the example above: a larger workpiece is allowed a wider tolerance at the same IT grade. The designer only has to note a grade code, and the machining side can convert it into the corresponding tolerance value based on the size.

03What can general CNC milling really achieve?

Back to the most practical question: how accurately can a well-maintained CNC mill reliably work day to day? The textbook rule of thumb can be grasped qualitatively like this:

• About ±0.02–0.05 mm: a range that general CNC milling under good conditions holds quite commonly and repeatably, suitable for most mechanical mating dimensions.
• The ±0.01 mm class: not impossible, but a "special arrangement"—it requires machine rigidity, a temperature-controlled environment, finishing toolpaths, sharp tools and measurement feedback all working together to hold reliably, with a clear rise in cost and time.
• Tighter tolerances (such as the sub-micron level or high IT grades): often beyond milling's reach, usually handed off to downstream finishing such as grinding or honing.

It must be stressed that the above is a qualitative rule of thumb, not a guaranteed value: what you can actually reach also depends on the size, material properties, feature type and clamping method (Kalpakjian & Schmid, 2020)^[1]. The same machine milling a hole in a compact steel part and milling the overall contour of a large aluminum plate can hold entirely different tolerances. So asking "how tight can this machine go" is itself incomplete—accuracy is the combined result of "machine + workpiece + method + environment."

04Five factors that affect accuracy

Why does the same drawing come out with different accuracy at different shops, on different days? The authoritative texts on metal-cutting mechanics and CNC design systematize the error sources; in practice the five most critical categories are (Altintas, 2012)^[2]:

Metal-cutting mechanics offers a particular reminder: feed, speed and depth of cut affect not only efficiency but also directly drive deflection, vibration and heat, which in turn determine the actual dimension and surface quality^[2]. That is why "chasing accuracy" often means "slow down, add finishing passes, control temperature, measure often"—each of which is a cost. For the details of the measurement side, see CNC inspection and measurement.

05Accuracy and cost: a non-linear curve

Add up the five factors from the previous section and you can see why the relationship between accuracy and cost is not a straight line but a curve that bends sharply upward. Going from a general tolerance to a medium one raises cost gently; but when the requirement approaches the limit of process capability, every further narrowing exacts a disproportionate price—slower cutting, more finishing passes, higher-end machines and fixtures, a temperature-controlled environment, denser 100% inspection, even an extra grinding operation.

Manufacturing engineering literature broadly describes this as a non-linear rise between accuracy and manufacturing cost: the closer you get to the process-capability boundary, the steeper the cost increase (Kalpakjian & Schmid, 2020)^[1]. This curve leads to a very practical conclusion—the reasonableness of a tolerance depends on how close it is to the process-capability limit. A tolerance in milling's comfort zone barely adds cost; a tolerance hugging the limit can change the quote and lead time dramatically even if it is only a few hundredths tighter.

06"You probably don't need 0.01": talking to the design side

Many high quotes actually stem from one habit: designers, to be safe, tighten every tolerance across the whole drawing, or carry over the tight tolerances of the previous revision without reviewing them again. The result is that a large number of non-mating dimensions are forced to be made with precision methods—cost goes up while function does not improve.

The healthy approach is to return to function itself and ask three questions of each dimension:

• What does it mate with? A dimension with no mating counterpart is usually fine with a general tolerance.
• What happens if it is a bit off? If it doesn't affect assembly, sealing or motion, it doesn't need a tight tolerance.
• Should it be expressed with a dimensional tolerance or a geometric tolerance? For position-type requirements, a position tolerance is often cheaper than forcing a tight dimensional tolerance—and closer to the intent.

Walk through these three questions and most drawings can be separated into a reasonable structure of "a few critical tight tolerances + a majority of general tolerances"—precisely the way of dimensioning that balances quality and price. For a complete drawing-dimensioning strategy for designers, see the tolerance and geometric tolerance (GD&T) guide.

07The role of AI dimensional cross-verification in accuracy management

The first checkpoint of accuracy management is in fact largely decided at the step of "the drawing being read correctly." If a tight tolerance is missed during modeling or quoting, or a general tolerance is mistaken for a precision requirement, no amount of machine accuracy afterward can make it up. AI's position here is clear—to speed up and verify the accuracy interpretation in the preparation stage, not to replace the final engineering judgment.

Specifically, when reading the drawing the AI recognizes the tolerance callouts and critical dimensions on it; then a second, independent AI cross-checks the generated 3D model dimensions against the original 2D drawing callouts one by one, proactively flagging inconsistencies. This dual "generate + independent check" architecture effectively reduces the human slip of "missing a tight tolerance" or "misreading a general tolerance as a precision one," so that quoting and process planning are built on the correct accuracy requirements.

But the boundary is equally clear: the final tolerance decision, the choice of datum, the measurement plan and whether downstream grinding is needed are still called by the professionals on the floor. AI is responsible for gathering the information and flagging the anomalies; the engineer is responsible for deciding "how this accuracy should be achieved and whether it is worth achieving." With that division of labor, accuracy management is both fast and stable.

08FAQ

09References

1. Kalpakjian, S., & Schmid, S. R. (2020). Manufacturing Engineering and Technology (8th ed.). Pearson.
2. Altintas, Y. (2012). Manufacturing Automation: Metal Cutting Mechanics, Machine Tool Vibrations, and CNC Design (2nd ed.). Cambridge University Press.
3. ISO 2768-1:1989. General tolerances — Tolerances for linear and angular dimensions without individual tolerance indications. International Organization for Standardization.