You send a part out for 3D scanning. The file comes back, opens fine in a viewer, and looks right on screen. Then you try to drop it into a CAD assembly, or send it to a machine shop for a quote, and everything stalls. The file is an STL — a mesh of triangles — and nothing downstream knows what to do with it. That gap between "a scan" and "a usable CAD model" is one of the most common misunderstandings in reverse engineering. Closing it is real engineering work, not a file conversion.
A structured-light or laser scanner doesn't capture "the part" as a design — it captures a cloud of XYZ points on the part's physical surface, often several million of them per scan pass. Software then stitches those points into a triangulated mesh: a shrink-wrap of flat triangular facets standing in for the real surface. Export that mesh and you get an STL (or OBJ, or PLY) — the format 3D Systems introduced in 1987 for stereolithography (hence the name), later backronymed as "Standard Tessellation Language." It's still the default handoff format for 3D printing and, by extension, for raw scan output.
An STL is missing almost everything a CAD file needs:
That's a legitimate output for some jobs. It is not a CAD model.
STEP (ISO 10303, "Standard for the Exchange of Product Model Data") is a boundary-representation, or B-rep, format. Instead of approximating a cylinder with a few hundred flat triangles, a STEP file defines it as an actual cylindrical surface — an exact mathematical definition, not a polygon guess. Edges are curves, faces are true surfaces, and it's the neutral format nearly every mechanical CAD package — SolidWorks, Creo, NX, Inventor, Fusion, CATIA — can read and write, which is why it's the default way to hand a solid model to someone on different software.
STEP files are generally editable as solids: add a fillet, revise a hole callout, extrude a boss. They can carry units and layer data, and in the newer AP242 flavor, PMI — dimensions and GD&T as structured, searchable data instead of a picture. That structured data is what lets a machine shop generate toolpaths, a mold builder calculate shrink and draft, or an engineer run a tolerance stack-up without re-measuring the physical part by hand.
| | STL (mesh) | STEP (B-rep solid) | |---|---|---| | Geometry | Triangular facets — approximate | Exact surfaces and curves | | Units | Not embedded | Embedded | | Editable in CAD | No feature tree | Yes, as solid features | | Tolerances / GD&T | Not supported | Supported (AP242) | | Typical origin | 3D scan, 3D-print export | Native CAD, reverse engineering | | Fits well for | Visual reference, replica prints | Machining, tooling, redesign, FEA |
Most CAD packages will import a mesh, let you wrap it in a NURBS skin, and spit out a file with a `.step` extension. That file will open. It still isn't a design — there's no feature intent behind it, and every scan artifact that was in the mesh is still there, just wrapped instead of removed. Real reverse engineering means a person deciding, region by region, whether a surface should be a plane, a cylinder, a specific fillet radius, or a free-form patch — and whether the part in your hand was supposed to be flat before it warped, or the bore was supposed to be round before it wore oval. Telling designed geometry apart from as-scanned condition is the actual engineering judgment call, and it's what makes the output trustworthy enough to hand to a machine shop.
Not every scan needs the full rebuild. Checking a rough fit, printing a one-off replica of a knob or bracket, or pulling a visual reference for a presentation — a cleaned-up mesh does the job, and a full parametric rebuild would be money spent for nothing. A mesh stops being enough the moment the file needs to carry a tolerance, feed an FEA study, go out for a mold quote, or get redesigned rather than just replicated.
Reverse-engineering price depends on part complexity, not a flat rate — true industry-wide, not just at one shop. A prismatic bracket with six flat faces and three bores might be an hour or two of modeling; a cast housing with blended fillets and free-form surfaces runs well past that, since every transition has to be built as a real surface, not traced. Ask any vendor for a fixed quote before they've seen the part, and geometry complexity is what they're pricing sight unseen.
At CADmore, short-range 3D scanning starts at $260 per part for the scan and mesh capture. Reverse engineering that mesh into a STEP model is priced separately — $300–$800 per part for 3D-printed parts, $800–$1,500 per part for injection-molded or CNC parts — and is part of the tens of thousands of projects submitted to CADmore, typically running 8-10 business days from scan to finished file. (Starting from a 2D drawing instead of a physical part? That's CAD conversion — same logic, no scan.)
Because CADmore builds validation prototypes in-house but doesn't run production itself, there's no incentive to sell more modeling than the part needs — the recommendation is whichever deliverable gets the part made correctly, on your manufacturer's equipment or through CADmore's partner network. If you're not sure whether your project needs a cleaned-up mesh or a full STEP rebuild, that's worth a real look at the part, not a guess: bring it (or your existing scan data) to a $100 consultation, or start a project and we'll scope it from there.
Not the way you'd edit a native model. Most CAD packages can import a mesh and wrap it as a dumb solid — a body with faces but no feature tree — which lets you measure it or add a few basic features on top, but you still can't drag a fillet radius or update a hole pattern the way you would on a part built from scratch. Real feature-based editability requires reverse engineering the mesh into a STEP or native CAD model.
Not by default — accuracy starts with the scan data either way. What STEP adds is exact mathematical surfaces instead of triangle approximations, so you're not stacking faceting error on top of scan error. The bigger accuracy question is usually design intent: a reverse-engineered STEP model can, and often should, correct for warp, wear, or scan noise that a raw mesh copies faithfully — so the STEP file may deliberately differ from the raw scan exactly where the physical part was out of spec.
Usually not for final tooling or tight-tolerance work. They need true solid geometry to generate toolpaths and calculate shrink and draft, and an STL can't carry tolerances or GD&T. Some shops will rough out a one-off prototype from a mesh, but any quote involving a dimensioned print or a fixed-price CNC or molding job is going to ask for a STEP or native CAD file.
Depends what you're doing with it. Ask for STL, OBJ, or PLY if you need a visual reference or a 3D-printed replica. Ask specifically for reverse engineering to STEP, and name the CAD package you'll open it in, if the part needs to be edited, dimensioned, machined, or tooled — that's a different scope of work than the scan capture itself, so say so when you request a quote.
At CADmore, reverse engineering typically runs 8-10 business days from scan to finished file, with part complexity — how much of the surface is simple prismatic geometry versus free-form — the main driver of where in that range a job lands.