What Is B-rep (Boundary Representation) in CAD?

You export a part to STEP for a supplier, they open it in a different CAD system, and something in the exchange goes wrong: a warning about invalid topology, a body that imports as surfaces instead of a solid, a healing prompt you have never seen before. Somewhere in that message is the term B-rep, and the software assumes you already know what it means.

This guide explains what B-rep (boundary representation) is, how its data structure works, why it matters in CAD modeling and manufacturing, and which file formats are built on it. If you work with STEP, IGES, or X_T files, you can also open and inspect them directly in a browser with the JLCDMP 3D Viewer, which reads these B-rep formats without a CAD installation.

The scope here is fundamentals and their practical consequences in CAD and manufacturing, not a modeling tutorial.

What Is B-rep (Boundary Representation)?

B-rep, short for boundary representation, describes a three-dimensional object by defining its boundary: the faces, edges, and vertices that separate what is inside the object from what is outside. The shape is never stored as a filled volume. It is stored as the skin around the volume, described precisely enough that the software can reason about the space the skin encloses.

Two kinds of information make that possible, and keeping them distinct is the key to understanding every B-rep behaviour that follows.

Topology is connectivity. It records which edges bound which face, which vertices sit at the ends of which edge, and which faces are adjacent to which. It carries no coordinates at all, only relationships.

Geometry is the mathematics. Each face points to an underlying surface equation: a plane, a cylinder, a cone, a torus, or a NURBS surface for anything freeform. Each edge points to a curve, and each vertex to a point in XYZ space.

A cylindrical hole illustrates the split neatly. The geometry says “cylindrical surface, radius 6.000 mm, axis along Z.” The topology says “this face is bounded by two circular edges, and it is adjacent to the top and bottom faces of the plate.” Change the radius and the geometry updates while the topology stays identical. That separation is what lets CAD systems modify shapes without rebuilding the entire description.

There is a practical wrinkle worth knowing about. Kernels do not insist on perfect mathematics; they work to tolerances, and each kernel and translator uses its own tolerance. Two faces whose edges differ by less than that tolerance are treated as joined, which is what makes filleting and booleans robust in the real world. It also means “exact” describes the surface equations rather than the stitching between them, and that distinction is behind most translation trouble.

One widespread claim needs correcting here. B-rep models are often described as watertight, but watertight is a property of solid bodies, not of B-rep as such. A solid body has a closed, fully connected boundary with no gaps. A sheet body, a single surface, a trimmed patch, an open shell, is also B-rep, and it is open by design. Surface modeling and solid modeling both run on the same underlying representation.

B-rep is the data architecture of the major geometric modeling kernels, Parasolid and ACIS among them. A kernel is the engine underneath the interface: it performs the booleans, filleting, and offsetting that turn user operations into valid geometry. When two CAD systems share a kernel, files move between them usually more cleanly. When they do not, translation is where problems appear.

How B-rep Works: Faces, Edges, and Vertices

A B-rep model is organised as a hierarchy, and the order matters more than most summaries admit.

At the top sits the body, the complete object. Below it are shells, each a connected set of faces, a solid with an internal cavity has two, one for the outer skin and one for the void. Below shells come faces, each a bounded region of an underlying surface. Each face is enclosed by one or more loops of edges, and each edge terminates at vertices.

So a face is a fundamental element, but it is not the top of the structure. Faces are contained within shells and bodies, not the other way around. Getting this right explains why a CAD system can report “two bodies” when you expected one, or why deleting a face can leave a valid open sheet rather than an error.

The three elements people meet in practice:

• Face - a trimmed piece of a surface. The underlying surface may be infinite, as a plane technically is; the loops of edges cut out the finite patch the part actually uses.
• Edge - in a closed solid, the boundary between exactly two faces, carrying its own curve equation: a line, a circular arc, or a NURBS curve for freeform transitions.
• Vertex - the endpoint where edges meet, stored as an exact XYZ coordinate.

Topology binds these into a coherent whole by recording adjacency. Because the model knows that face A and face B share edge E, it can determine that the boundary is closed, calculate the enclosed volume, and offset the whole thing by 2 mm without losing track of which neighbour belongs to which.

This is also where imports fail. If a translated edge leaves a gap of 0.01 mm between two faces, the geometry still looks correct on screen, but the topology no longer closes, and the receiving kernel reports a shell that cannot become a solid. The geometry was fine; the connectivity was not.

Why B-rep Matters in CAD Modeling

Precision. Because surfaces are stored as equations, a radius is exactly the radius specified, at every point, at any zoom level. There is no resolution setting and no deviation to tune. What you see on screen, in CAD or in a web viewer, is a tessellated display mesh; the stored surfaces stay exact.

Engineering analysis. Volume, mass, centre of gravity, and moments of inertia come from integrating over exact surfaces, so the numbers are accurate rather than approximated. FEA meshers also work far better from clean B-rep input, because they can refine the mesh against the true surface instead of against a fixed set of triangles.

Parametric editing happens one level up. This is the point that causes the most confusion. Your feature tree, extrude, fillet, pattern, lives in the CAD application. The kernel evaluates that tree and outputs the resulting B-rep. When you export to STEP, IGES, or X_T, the output travels and the recipe stays behind. The recipient gets accurate geometry with no editable dimensions, which is exactly what “dumb solid” means. Native formats such as SLDPRT or CATPart keep the feature history alongside the geometry, which is why they preserve editability and exchange formats do not.

Feature recognition. CAM and DFM tools read topology to identify machinable features, a cylindrical face with two circular edges reads as a hole, a set of parallel faces reads as a pocket. That inference is only possible because the connectivity is there to be queried, which is why CNC and sheet-metal quoting works best from STEP files, while 3D printing quotes also accept mesh files such as STL.

Manufacturability. CAM systems consume B-rep directly. Toolpath generation needs to know a pocket floor is planar and a boss is cylindrical, not that both are collections of triangles. Mold design, sheet metal unfolding, and CMM inspection routines all depend on the same exactness. When a model crosses kernels, Parasolid to ACIS, say small tolerance differences can leave gaps that need healing before any toolpath will generate.

B-rep vs Mesh: Key Differences

Both describe a 3D shape; they answer different questions. A mesh asks what the surface looks like. B-rep asks what the surface is.

When to choose which. If your downstream work involves CNC programming, FEA, parametric iteration, or precise volume and mass figures, you need B-rep data. If the goal is visualisation, web viewing, 3D printing slicing, or real-time rendering, a mesh is sufficient and considerably faster to load. Plenty of workflows use both: B-rep for design and analysis, mesh for sharing and display.

The file size row deserves a concrete example. A turned shaft with a few diameters and a fillet might be a few hundred kilobytes as STEP, because each surface is one equation and a handful of parameters. Exported as a mesh fine enough to machine from, the same part can run to tens of megabytes, and it still will not be exact. For flat-faced parts the comparison narrows, and for heavily faceted scan data the mesh can be the smaller file.

Common CAD File Formats That Use B-rep

B-rep is the underlying architecture of most formats that move engineering data between systems. They differ in how completely they carry the topology. For mesh and print formats, see our guide to 3D printing file formats.

If you need to open any of these without the originating software, the JLCDMP 3D Viewer handles most of them in a browser, including measuring distances, angles, radii, surface area and volume. Check its homepage for the current list of supported file types.

FAQ about B-rep (Boundary Representation) in CAD

Conclusion: B-rep (Boundary Representation) in CAD

B-rep is a foundational data architecture rather than a feature you switch on. It underpins the exchange formats engineers use daily, STEP and IGES among them, and it is the shared basis of mainstream kernels including Parasolid and ACIS. Knowing how it separates topology from geometry explains most of what distinguishes these formats from one another.

It also explains the everyday frictions: why an imported model has no feature tree, why a translated part sometimes needs healing, why a mesh cannot simply be turned back into a solid. None of those are faults in the software. They follow directly from what each representation was designed to do.

Whether you are working in STEP, IGES, or a native CAD format, B-rep is the architecture running quietly underneath, keeping your model precise, manufacturable, and ready for analysis.