There Is No Such Thing as an Exact Dimension

A CAD model says 40.00 mm. Nothing you make will be 40.00 mm. It will be 39.86 or 40.11, and whether that is a defect or a non-issue depends entirely on what the feature does. The useful question is never "is the part accurate" but "which of these dimensions had a job, and did it do it."

Founders inspecting a first prototype tend to measure everything with calipers and get alarmed at the first tenth-of-a-millimeter deviation. Meanwhile the bore that actually decides whether the shaft spins goes unmeasured. Sorting features into critical and non-critical before parts arrive is the whole discipline.

Typical Deviation by Process

These are the numbers a competent shop will hold without a special request. Anything tighter costs more and needs to be called out.

  • FDM. Roughly plus or minus 0.5 percent with a floor around 0.010 to 0.020 in (0.25 to 0.5 mm). Holes come out undersized by 0.004 to 0.012 in (0.1 to 0.3 mm) because of extrusion overshoot on inside curves. Z height is the most repeatable axis; XY suffers from shrinkage and warp.
  • SLA and DLP resin. Roughly plus or minus 0.005 to 0.010 in (0.13 to 0.25 mm) on small parts. Excellent on fine detail, but green parts continue to cure and shrink for days, and thin sections sag under their own weight before post-cure.
  • SLS nylon. Roughly plus or minus 0.3 percent with a floor near 0.012 in (0.3 mm). Large flat surfaces bow from thermal gradients in the powder bed. See SLS nylon 3D printing for where the process shines despite that.
  • Urethane casting. Roughly plus or minus 0.010 in (0.25 mm) plus accumulated shrinkage, and the silicone tool degrades over 20 to 25 pulls, so part 25 is measurably different from part 1. Details in urethane casting for low-volume production.
  • CNC machining. Roughly plus or minus 0.005 in (0.13 mm) as standard shop tolerance, plus or minus 0.001 in (0.025 mm) on called-out features, tighter with grinding. This is the only prototype process where you can simply specify what you need, as covered in CNC machining for prototypes.

Two things get lost in these tables. First, tolerance is not the same as repeatability: a printer that is consistently 0.3 mm undersize is easy to compensate for, while one that varies randomly by 0.3 mm is not. Second, the number applies per feature, not per part. Five stacked features can drift five times as far, which is why tolerance stack-up analysis matters even on a prototype.

Where Accuracy Actually Matters

On a normal product, fewer than ten percent of dimensions are critical. They fall into recognizable groups.

Mating interfaces. Bores for bearings and shafts, gland grooves for O-rings, press-fit diameters, dovetails, and rail spacings. These decide whether the mechanism works, and a tenth of a millimeter here is a real failure.

Off-the-shelf component pockets. The slot for a purchased LCD, the footprint for a PCB and its mounting holes, the cavity for a battery cell, the seat for a purchased pump. These are set by someone else's drawing and you have zero room to negotiate.

Sealing surfaces. Anything that has to hold water or air. Flatness and surface finish matter as much as the linear dimension.

Regulated or ergonomic dimensions. A grip diameter validated with users, a guard opening set by a safety standard, an insertion depth on a medical device.

Everything else, meaning wall thicknesses, cosmetic radii, rib heights, overall envelope, and screw boss positions with clearance holes, can float by half a millimeter without anyone noticing. Marking only the critical set on the drawing is also the fastest way to cut a quote, which is one reason GD&T basics pays for itself even outside production.

The Part Keeps Moving After You Measure It

Dimensional deviation is not fixed at the moment the part comes off the machine.

Nylon absorbs moisture and can grow 0.2 to 0.5 percent between a dry printer chamber and a humid room, which on a 6 in (150 mm) part is up to 0.030 in (0.75 mm). ABS and PLA relax residual stress over the first week and warp slightly. Resin parts shrink for 24 to 72 hours after post-cure. Aluminum expands about 0.0000128 per degree F, so a 10 in part measured in a 60 F shop and used in a 100 F enclosure grows about 0.005 in.

Practical consequences: measure parts at least a day after they are made, not the hour they arrive; measure at a stated temperature and write it down; and if a part will live outdoors or in a hot enclosure, check the fit at temperature rather than on the bench.

Designing a Model That Tolerates the Error

The cheapest response to prototype inaccuracy is to design around it instead of chasing it.

  • Oversize holes and ream or drill them to size after printing. A reamed hole in a printed part is far more accurate than a printed one.
  • Use slots and elongated holes so position error becomes adjustment range rather than scrap.
  • Insert a machined bushing, sleeve, or plate wherever a precise diameter or flat is required, and let the printed part hold it loosely.
  • Add shim pads at every stacked interface so you can tune the assembly by 0.005 in increments.
  • Apply a scale factor when you know the direction of the error. If your printer runs 0.4 percent small in XY, model at 100.4 percent rather than filing every part.
  • Keep long thin features short. Warp scales with length, so a 10 in flat panel will bow where two 5 in panels joined by a bracket will not.

What to Tell the Shop

Send a STEP file plus a simple drawing that marks three to six critical dimensions with the tolerance you actually need, states the datum surfaces those dimensions reference, and names the material. Ask whether the quoted process can hold them; a good shop will tell you when it cannot and suggest a secondary operation. Do not put plus or minus 0.002 in on every dimension of a printed part. It reads as inexperience and the shop will either ignore it or price the whole part as if it were machined.

Get the Critical Dimensions Called Out Correctly

Projects House prepares prototype drawings that separate the handful of dimensions that decide function from the ones that only have to look right, then picks the process and the secondary operations to hit them. Send your CAD and what the model has to prove through our contact form.