Do CNC Machines Use STL?

• 1
  What an STL File Actually Contains
• 2
  How an STL Becomes Toolpaths
• 3
  When an STL Is Still the Right File
• 4
  What the Wrong Format Costs You
• 5
  How to Prepare Files for a CNC Quote
• 6
  STL vs STEP for CNC Machining

An STL file stores one thing: a triangle mesh that approximates the outer surface of a solid. There are no units, no datums, no feature names, no thread callouts, no tolerances. Just a list of vertices and normals. Two files can describe the same bracket and differ by a factor of 25.4, and nothing inside either file tells you which one is right.

That is fine for a slicer, because 3D printing is a surface-driven process. The machine deposits material along a path derived from the mesh. A CNC machine is different. The cutting tool follows a commanded path in machine coordinates, referenced to a work origin, with a defined tool diameter and a stock allowance. Between the mesh and the motion there is a whole preparation step.

So the honest answer to "do CNC machines use STL" is that the machine never sees it. A CAM programmer reads the file, converts the geometry into toolpaths, posts G-code, and the control executes that. The STL is an input to software, not to hardware. Any claim that a mill or lathe consumes STL natively is describing the front end of a CAM chain, not the machine.

CAM software can import an STL, but it treats the model as dumb geometry. Feature recognition fails, so every pocket, hole, fillet and chamfer has to be rebuilt or selected manually. On a part with twenty holes that is an afternoon. On the same part supplied as STEP, the holes arrive as cylinders with axes and diameters, and the programmer assigns drill cycles in minutes.

Chord tolerance is the second problem. A curved surface exported at a coarse setting becomes a faceted shell, and the CAM system cuts the facets. The tool leaves visible steps. Tighten the export and the file grows fast, which slows regeneration and can crash older CAM seats. You are paying for tessellation you never wanted.

Watertightness matters too. Slicers repair small gaps on the fly. CAM usually will not. A mesh with flipped normals or open edges produces toolpaths that plunge through the part or skip a face entirely. Repair tools exist, but every repair is time on a quote you are waiting for.

Once the geometry is clean, the normal CAM chain applies. Roughing clears the bulk with a 12–20 mm end mill, semi-finishing follows, then finishing with a 6 mm or 3 mm cutter at Ra 0.8–1.6 μm. Toolpath tolerances start around 0.01 mm and tighten for the finishing pass. None of those numbers come from the STL. They come from the shop's process plan.

There are real cases where STL is the practical choice. If the part is an organic, sculpted shape with no prismatic features, the CAD kernel behind it may already be mesh-based. Sending the mesh avoids a lossy conversion to a solid that CAM then has to reverse. Lattice and topology-optimized geometries often fall into this group.

STL is also fine for an early feasibility check. A shop can slice the mesh, look at wall thickness and tool access, and tell you whether a 3 mm cutter reaches the internal corner before you commit to a full DFM report. That is a rough check, not a production definition.

A third case: the part is being 3D printed first and machined later. The same mesh drives the printer, and the machined version gets its own solid model. Keeping the two separate avoids the trap of machining a print file that was never dimensioned.

Outside those cases, the mesh path costs time. For a bracket, a housing, a manifold or a shaft, the features are the part. Strip them out and you have handed the shop a modeling job plus a machining job, and only one of those was quoted.

The first cost is time in the quote queue. A STEP file with clean features usually returns a quotation and DFM analysis within 12 hours. A dense, repaired mesh takes longer because someone has to rebuild reference geometry before a price means anything.

The second cost is dimensional risk. If the mesh has no units, a programmer who assumes millimeters on a part designed in inches gets a scale error of 25.4×. That is not a subtle failure. It shows up as a scrapped blank and a lost week.

The third cost is tolerance drift on curved surfaces. A faceted boss machined to the mesh will measure smaller than the nominal cylinder. On a press fit that is a rejected assembly. On a cosmetic cover it may never be noticed until the customer runs a CMM check.

For regulated work the format question is not optional. Aerospace, automotive and medical programs need traceable dimensions that match the drawing. A mesh cannot carry that. This is one reason our shops work from STEP or native CAD, inspect 100% before shipment, and hold ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022.

Export STEP (AP214) from your CAD system and include the drawing or a PDF with critical dimensions, datums and GD&T. If a feature is cosmetic, say so. If a hole is a press fit, call out the tolerance. Those notes remove assumptions before the programmer makes them.

If you only have an STL, send it plus the original CAD if any exists. Add a note on units and the intended function of each critical face. A programmer can rebuild the solid faster when the intent is clear, and the quote will be closer to final.

Keep the model at nominal size with sharp corners. Do not pre-apply chamfers for tool radius or add draft for machining. The shop decides cutter compensation. Pre-modified geometry hides intent and makes the DFM review argue with your model instead of your design.

For assemblies, send the individual parts, not just the assembly. A shop quoting a single machined component needs the component's own coordinate system. Also state the material from our stocked list, such as 6061-T6, 304 stainless, 17-4PH or Ti-6Al-4V, because machinability changes the process plan and the price.