Gravel road categories: the Cat 1–4 ISGG scale and what tyres each needs
"Gravel" covers everything from a hard-packed farm road you could ride on 25s to an unmaintained forest track with rock gardens in it. One scale has become the sport's de-facto vocabulary for telling them apart — four categories, each with a tyre-width recommendation. Here it is, plus how it lines up with the tags in OpenStreetMap and what each class costs you in rolling resistance.
The ISGG: four categories, and a tyre for each
The Industry Standard Guide to Gravel was created by Neil Shirley, an ex-pro road racer who went on to edit Road Bike Action. It grades a gravel road on maintenance state and debris, and attaches a tyre width to each grade:
| Class | Name | What it is | Tyre |
|---|---|---|---|
| Cat 1 | Smooth with sparse gravel | "Smooth, well maintained dirt roads that have either very small gravel chunks, or none at all" — "hard packed, offering little more difficulty than riding on tarmac" | 25–28 mm |
| Cat 2 | Gravel outside tracks | "potholes, washboard and probably loose, blown-out corners. There is likely to be gravel outside of the main tire tracks" | 28–32 mm |
| Cat 3 | Chunky gravel | "Infrequently maintained roads [with] exposed rocks, tire eating rain ruts, sand bogs and any number of other unexpected challenges" | 33–38 mm |
| Cat 4 | Unmaintained forest roads | "deep ruts, rock gardens, and potential landslides" | 38–42 mm |
What lifts this above one blogger's private scheme is that the industry picked it up. SILCA's tyre-pressure calculator rates surfaces on the same scale and refers to "category 4 gravel which is the worst surface in our calculator"; their gravel tyre-testing write-ups talk about Cat 1 and Cat 2 pressures and test courses. ENVE's own gravel guide describes the same four-step progression, right down to the exposed rocks, rain ruts and sand bogs at the harder end.
The caveat, stated plainly: "Industry Standard" is the scheme's own name, not a designation by any governing body. The UCI, which now runs a Gravel World Championships, publishes no surface grading. There is no ISO or CEN standard, and no published rolling-resistance figure per class. Velo ran a piece asking how one would even define off-road surfaces — its existence is itself evidence the question is unsettled in the sport.
Grading a route rather than a road
The route-publishing site Graveleur uses three grades instead of four, and its useful idea is a different one: it grades a whole route by its worst surface, which is what a rider actually needs in order to pack.
- Grade 1 — smooth compacted dirt or gravel, "suitable for any bike", small ruts or loose patches.
- Grade 2 — loose and uneven gravel, smooth singletrack; erosion, water bars, sandy sections.
- Grade 3 — technical: washed-out fire roads, loose sand, rock gardens, hike-a-bike; "probably better suited for MTB only".
It collapses ISGG's Cat 1 and Cat 2 into its Grade 1 and is otherwise the same axis.
The OpenStreetMap vocabularies, and why they can't do the job alone
Two OSM tags grade the same thing from a mapping rather than a riding perspective.
tracktype grades firmness, not material
| Value | Definition (OSM wiki) |
|---|---|
grade1 | "Solid. Usually a paved surface" |
grade2 | "Mostly solid. Usually an unpaved track with surface of gravel mixed with a varying amount of sand, silt, and clay" |
grade3 | "Even mixture of hard and soft materials" |
grade4 | "Mostly soft … prominently with soil/sand/grass, but with some hard or compacted materials mixed in" |
grade5 | "Soft. An unimproved track lacking hard materials, uncompacted" |
smoothness grades usability by vehicle
This is the only one of these vocabularies that names bicycles explicitly, which makes it the
best cross-walk to ISGG: excellent (all vehicles), good (racing
bikes), intermediate (city bikes), bad (trekking bikes),
very_bad (high-clearance cars), horrible (heavy off-road),
very_horrible (tractors, ATVs, mountain bikes), impassable
(nothing wheeled).
Coverage is the problem, not the vocabulary. In OSM, highway is
tagged on roughly 265 million ways but surface on only about 79 million, and
smoothness is far rarer than either. So OSM can inform a surface class
for a stretch of road, but it cannot supply one for a whole route.
An approximate cross-walk
This table is ours, and approximate for an honest reason: the three axes measure different things — debris, firmness, and usability by vehicle — and no source aligns them.
| ISGG | Graveleur | OSM tracktype | OSM smoothness |
|---|---|---|---|
| Cat 1 | Grade 1 | grade1–grade2 | good / intermediate |
| Cat 2 | Grade 1–2 | grade2–grade3 | intermediate / bad |
| Cat 3 | Grade 2–3 | grade3–grade4 | bad / very_bad |
| Cat 4 | Grade 3 | grade4–grade5 | very_bad / horrible |
What the surface actually costs you
Rolling resistance — sourced, and coarse
Published rolling-resistance coefficients (Crr) for a bicycle road tyre on tarmac span 0.002–0.007, commonly 0.004–0.005. Off tarmac, the sourced guidance is a surface additive rather than a per-class table: a dirt or gravel fraction is modelled by adding about 0.004 to the tyre's tarmac Crr — roughly a doubling — and extended thick sand needs an additive nearer 0.028. Coarser summaries land in the same place: asphalt ≈ 0.004, gravel or packed dirt ≈ 0.008, mountain-bike trail ≈ 0.015, with off-road surfaces able to exceed 0.012.
tracé's pacing model prices five surfaces. The endpoints of that column are sourced; the two middle-and-beyond values are not, and the app says so:
| Class | Crr | Where the number comes from |
|---|---|---|
| Road (paved) | 0.005 | Inside the sourced tarmac band |
| Cat 1 | 0.008 | Sourced endpoint — "gravel or packed dirt" |
| Cat 2 | 0.011 | Ours — interpolated |
| Cat 3 | 0.015 | Sourced endpoint — "mountain-bike trail" |
| Cat 4 | 0.020 | Ours — extrapolated, short of the 0.028 sand figure |
Aerodynamic drag — unchanged by surface
No source suggests drag area differs by surface, and it does not: a rider's frontal area is their position, not the road. A gravel rider is more often on the hoods than in a tuck, but that is an argument about position, not about the surface.
Descending speed — not sourced at all
Nothing was found. No governing body, manufacturer or published model defines a descending speed limit for any surface, on tarmac or off it. tracé's pacing model does impose a braking ceiling that falls as the surface gets rougher, and every one of those numbers is ours — a flat 5 km/h step per class at both ends of the ladder:
| Class | Braking ceiling (shallow / steep descent) | Provenance |
|---|---|---|
| Road | 50 / 35 km/h | Ours — unchanged from the model before gravel existed |
| Cat 1 | 45 / 30 km/h | Ours |
| Cat 2 | 40 / 25 km/h | Ours |
| Cat 3 | 35 / 20 km/h | Ours |
| Cat 4 | 30 / 15 km/h | Ours |
What is defensible without a citation is the direction and the mechanism, not the figures. A descending rider is limited by brakes, corners, sightlines and traction, and on loose gravel all four get worse: braking distance grows, cornering grip falls, and a rider who cannot read the surface texture ahead rides within a shorter sight line. The ISGG definitions describe that mechanism themselves — "loose, blown-out corners" at Cat 2, "tire eating rain ruts, sand bogs" at Cat 3 — which is qualitative support for a falling ceiling, but not a number for one. Treat any such figure, ours included, as a modelling choice.
How tracé handles surfaces, and what leaves your browser
The surface class is the rider's declaration, not a derived value — a GPX file carries no surface information at all, and OSM tag coverage cannot support deriving one route-wide. You tell the app what you are riding; it changes exactly two things per class, the rolling resistance and the braking ceiling. Not drag area, for the reason above.
The optional surface layer can refine that declaration stretch by stretch using OpenStreetMap data, and it is worth being straight about what that involves. Surface data comes from OpenStreetMap, fetched over the network through the app's one piece of server-side code: a stateless relay for OpenStreetMap Overpass queries, which exists only because browsers are no longer permitted to call Overpass directly. It stores nothing, it runs only when you switch the layer on, and what it sees is bounding-box coordinates — never your file. Route analysis itself is entirely in-browser, with no upload and no account. If you never enable the layer, no request is made.
Where OSM and your declaration disagree, the declaration usually wins: if you said road, the app treats it as road, because a partially-tagged guess should not override a rider who knows. A stretch tagged as paved on a gravel route is priced as pavement. A stretch tagged simply "unpaved" is not silently stepped one class rougher — it genuinely is rougher than gravel, but no source supports a step of any particular size, and inventing one would be a threshold nobody could label.
Sources
ISGG categories, definitions and tyre widths: The Unpaved Hub. Industry adoption of the scale: SILCA — tyre-pressure calculator explained, SILCA — tyre-pressure testing for Unbound Gravel, ENVE's gravel guide. Three-grade route grading and worst-surface-wins: Graveleur. OSM vocabularies: Key:tracktype, Key:smoothness. Rolling resistance: BestBikeSplit, Zwift Insider — Crr, Zwift Insider — Crr charts, Bicycle Rolling Resistance — the test, and a comparison of MTB tyre rolling resistance across diameters and surface conditions.
Tell tracé what surface your route is and get a pacing estimate that prices it — no upload, no account.
Open tracé