Tolerance Stack-Up in CNC Machining: Why Tighter Isn’t Always Better

“Tighter tolerance is better.” It’s the most common assumption in custom manufacturing, and it’s wrong. Tighter tolerances cost more, take longer, and — when applied carelessly — can actually make assemblies worse. Tolerance stack-up is the engineering discipline that tells you what tolerance each feature actually needs to make the assembly work. It’s how you save 30% on machining costs without sacrificing a single millimeter of function.

This is a practical primer for designers and engineers who want to specify tolerances based on the assembly’s real needs, not on what feels safe.

What Stack-Up Actually Is

Tolerance stack-up is the cumulative effect of individual feature tolerances on a final assembly dimension. If you have five parts in a row, each with a length tolerance of ±0.005″, the total stack on the assembled length isn’t ±0.005″ — it’s somewhere between ±0.011″ (statistical) and ±0.025″ (worst-case), depending on how you do the math.

The implication: you can’t pick tolerances on individual features in isolation. You have to know what assembly dimensions matter and work backwards to the feature tolerances that satisfy them. The good news is that this almost always lets you loosen non-critical features without affecting function.

Two Methods, Two Answers

Worst-case stack-up

Add up every individual tolerance assuming all parts are simultaneously at their worst-case extremes. This is the most conservative approach and the one most engineers default to. The math is simple — sum the tolerances of every feature in the chain.

Worst-case is appropriate for safety-critical assemblies, low-volume parts where statistical sampling doesn’t apply, and any case where a single failed assembly is unacceptable. The downside: it’s typically 2–3x more conservative than reality, which means you’re tightening tolerances that don’t need to be tightened.

Statistical (RSS) stack-up

Root-sum-square the individual tolerances. The math is based on the statistical reality that not every part will be at its worst-case dimension simultaneously — the variations partially cancel each other out. RSS gives you a tolerance band that 99.7% of assemblies will fall within, which is acceptable for most consumer and industrial products.

Use RSS when you have production volumes high enough that statistical sampling applies and individual assembly failures are recoverable (rework, replace, sort). It typically lets you loosen feature tolerances by 30–50% versus worst-case.

A Concrete Example

Consider a shaft with three bearings on it. Each bearing seat has its own diameter tolerance. The customer wants the assembled bearing-to-bearing alignment within ±0.003″.

Worst-case approach: each of the three bearing seats needs ±0.001″ tolerance, plus tighter tolerances on the housing bores. Multiple grinding operations. High cost.

Statistical approach (with proper SPC and reasonable production volume): each bearing seat can hold ±0.002″, and the assembly will still meet the ±0.003″ alignment 99.7% of the time. The features that fall outside that band can be sorted at assembly. Single turning operation with light grinding only on critical surfaces. Substantial cost reduction.

Same final assembly performance. Very different manufacturing cost.

When Tighter Tolerance Hurts

1. Manufacturing cost goes up exponentially

The cost-tolerance curve is brutal. Going from ±0.005″ to ±0.002″ on a typical machined feature might add 30% to the cost. Going from ±0.002″ to ±0.0005″ can double it. Going from ±0.0005″ to ±0.0001″ can triple it again. Most of that cost increase delivers no functional benefit if the assembly didn’t need the precision.

2. Inspection time scales with tolerance

Tight tolerances need precision metrology — CMM, optical comparator, air gauges. Each measurement takes longer, requires more skill, and is more sensitive to environmental conditions like temperature and vibration. A drawing full of tight tolerances is one that takes twice as long to inspect, and inspection is rarely free.

3. Press fits become unmanageable

For interference and transition fits, the assembly force is exquisitely sensitive to the actual dimensions. Specify too tight a hole tolerance and you can end up with parts that won’t press together, parts that gall during installation, or parts where the bore is so close to the shaft that thermal cycling distorts the assembly.

4. You lose flexibility for repair and replacement

An assembly designed for ±0.0001″ matched components is one where field replacement is impossible. The replacement part must be custom-matched to the existing assembly. For products that need to be repairable in the field, looser tolerances and proper functional design are almost always worth the trade-off.

When Tighter Tolerance Genuinely Helps

Tightening tolerances is the right answer when:

The assembly has functional requirements that genuinely need precision — bearing fits, sealing surfaces, optical alignment
The cost of an out-of-spec assembly exceeds the cost of tighter manufacturing — aerospace, medical, safety-critical
Production volumes are too low for statistical sampling to apply
The assembly cannot be sorted or reworked
You’ve already loosened everything else and the math still doesn’t work

A Practical Stack-Up Workflow

Identify the critical assembly dimensions — what gaps, alignments, or fits actually matter for function?
Map the dimension chain — for each critical assembly dimension, list every feature on every part that contributes
Set initial tolerances — start with what’s economical to manufacture (typically ±0.005″ for milled features, ±0.002″ for ground or turned)
Run the worst-case math — does it fit?
If not, run RSS — does it fit statistically? Is statistical acceptable for this product?
If still not, identify the critical contributors — which features have the biggest leverage? Tighten those first.
Loosen everything else — once you’ve found the features that actually need tight tolerance, the rest can usually open up
Verify with a sample run — measure real parts and confirm the math

The Designer-Manufacturer Conversation

The fastest way to optimize tolerances is to talk to your manufacturer before finalizing the drawing. A good machine shop can tell you immediately which tolerances are economical to hit, which ones drive setup time, and which ones are essentially free because they fall out of the natural process capability anyway.

For example: turning a feature to ±0.002″ on a CNC lathe is essentially free if the part was already going to be turned. Holding ±0.0005″ on the same feature requires grinding — a separate operation, separate fixture, separate inspection. Same drawing, different process route, very different cost.

Common Stack-Up Mistakes

Forgetting fastener clearance

Bolt holes have clearance, and that clearance contributes to the stack. A pattern of four bolts holding a part in position has more positional flexibility than the GD&T position tolerance alone suggests. Use that flexibility — don’t fight it.

Ignoring thermal expansion

Steel grows about 0.000007 inches per inch per °F. A 12″ assembly seeing a 100°F swing expands 0.008″. If your stack-up doesn’t account for the operating temperature range, you’ll get assemblies that work in the QC lab and fail in the field.

Treating GD&T position tolerance as plus-or-minus

A position tolerance of ⌀0.010 isn’t ±0.010 in X and ±0.010 in Y — it’s a circular zone with a 0.010 diameter. For stack-up math, that’s about ±0.005 in any single direction. Mixing the two conventions gives you a stack that’s twice as tight as the drawing actually specifies.

Skipping the bonus tolerance

When position tolerance is specified at MMC, you get bonus tolerance as the feature size moves away from MMC. That bonus is real and statistical stack-up should include it. Worst-case stack-ups often miss this and end up with overly conservative results.

Where We Come In

Tolerance optimization is one of the highest-ROI engineering reviews you can do on a drawing. We routinely cut machining costs 20–40% on customer drawings just by working through the stack-up math and identifying which features actually need precision and which were carrying conservative defaults. Send us your drawing and we’ll come back with a quote and a tolerance review at the same time — no charge for the review.

DFW Machine