FTP vs Critical Power: Which Threshold Should You Train By?
FTP and Critical Power both estimate your sustainable cycling effort, but they are built differently. Here is how they diverge, how to test each, and which to use.
You did a 20-minute test, multiplied by 0.95, and your head unit now shows an FTP. Then a friend on a different platform tells you your Critical Power is 12 watts lower, and that you also have a W' of 19 kilojoules, whatever that means. Same legs, same power meter, two different numbers claiming to describe the same thing.
They are not describing the same thing. FTP and Critical Power both try to answer one question, what is the hardest steady effort you can hold, but they come from different places and behave differently when you push on them. Understanding where they agree and where they split apart is the difference between pacing a climb correctly and blowing up two-thirds of the way up.
What FTP actually is
Functional Threshold Power was popularized by Andrew Coggan and Hunter Allen in their book Training and Racing with a Power Meter. The definition most people use is the highest power you can sustain in a quasi-steady state for approximately one hour.
The reason almost nobody tests it with an actual one-hour effort is that a true 60-minute maximal effort is brutal and hard to pace. So the practical shortcut became the 20-minute test: ride all out for 20 minutes, take 95% of that average power, and call it your FTP. The 5% haircut is an approximation of the gap between 20-minute and 60-minute power for a typical trained rider.
That word "approximation" is doing a lot of work. FTP is a single number produced by a single protocol with a fixed correction factor. It is robust, repeatable, and good enough for most training. But it bundles your aerobic ceiling and your anaerobic contribution into one figure, and it assumes the 95% relationship holds for you specifically. For a rider with a big anaerobic engine, a strong 20-minute test can inflate FTP above what they can truly hold for an hour.
FTP's biggest strength is the ecosystem around it. Power zones, the structured workout files you download, and most analytics platforms all key off FTP. If you serialize a workout as a fraction of FTP, it travels cleanly between your indoor trainer and your head unit. That standardization is why FTP is not going anywhere.
What Critical Power actually is
Critical Power has deeper roots in exercise physiology. The concept goes back to Monod and Scherrer in 1965, who modeled the relationship between the intensity of a muscular task and the time you can hold it. Applied to cycling, the model says that your sustainable power and your time to exhaustion follow a hyperbolic relationship.
The cleanest way to see it is the two-parameter model:
Power = (W' / time) + CP
CP is the horizontal asymptote, the power the curve flattens toward as duration gets long. It represents the boundary between two physiological worlds: below CP, you reach a metabolic steady state and can ride for a long time; above CP, you are in a non-steady state where fatigue accumulates and exhaustion is only a matter of time.
W' (W-prime) is the curvature constant, the finite chunk of work you can do above CP before you run out. It is measured in kilojoules and behaves like a battery. Ride above CP and W' drains. Drop below CP and it slowly recharges. A 2016 review by Poole, Burnley, Vanhatalo, and Rossiter in Medicine and Science in Sports and Exercise laid out the strong physiological case that CP marks a genuine threshold above which muscle homeostasis cannot be maintained.
This two-parameter structure is what makes CP more than a fancier FTP. It does not just give you a sustainable power, it gives you a model of how much you can spend above that power and for how long. Jones and Vanhatalo, writing in Sports Medicine in 2017, argued that this is exactly why the CP model describes intermittent, surge-heavy efforts better than a single threshold number can.
Where the two numbers diverge
Here is the part that confuses people. FTP and Critical Power are often close, sometimes within a handful of watts, but they are not the same and the gap is not random.
Test protocol drives most of the difference. FTP from a single 20-minute test leans on that 95% correction. CP from short efforts, say a 3-minute and a 12-minute, is anchored partly by your anaerobic power and can land higher. CP from longer efforts tends to land lower and closer to true hour power. Change the durations you test and you move the CP estimate.
Anaerobic capacity skews the relationship. A track sprinter and a time triallist can post the same 20-minute power and therefore the same FTP, while having very different Critical Power. The sprinter's huge W' inflates short efforts, which can pull a short-protocol CP estimate upward even though their genuinely sustainable power is lower. FTP cannot see this distinction at all because it has no W' term.
Steady state is defined differently. Research on the maximal metabolic steady state suggests CP often sits a little above the power most riders can hold for a full hour, while FTP is explicitly anchored to roughly one-hour power. So depending on your physiology, CP can read above or below FTP. Treating them as interchangeable is where pacing errors creep in.
A quick worked example makes the gap concrete. Two riders both average 300 watts for 20 minutes, so both get an FTP of 285 watts after the 5% haircut. Rider A is a steady diesel with a small W' of 12 kJ; their CP, fit from a 3-minute and a 12-minute effort, comes out around 282 watts, almost identical to FTP. Rider B is punchy with a 26 kJ W'; their short efforts are inflated by all that anaerobic capacity, so a short-protocol CP reads 296 watts even though their genuine hour power is lower than rider A's. Same FTP, very different engines, and only the CP-W' model can tell them apart.
For practical purposes, expect them to be in the same neighborhood but plan for a 10 to 30 watt spread. The point is not to crown a winner. It is to know which model you are using before you build workouts or pace a race off it.
How to test each one
Testing FTP. The standard is a single 20-minute maximal effort after a thorough warm-up, ideally on a steady road or a trainer in a stable environment. Take the average power, multiply by 0.95, and that is your FTP. Some riders prefer the 8-minute protocol, taking 90% of a single 8-minute effort, but it is noisier and tends to overestimate. Ramp tests, which raise power in steps until failure, estimate FTP from your final minute and are convenient on smart trainers, though they read high for riders with strong neuromuscular power.
Testing Critical Power. You need at least two maximal efforts of different durations, performed fresh. A common pairing is around 3 minutes and 12 minutes, sometimes with a 5 to 7 minute effort added for a better fit. Do them on separate days, or with very long recovery between, because a fatigued second effort corrupts the model. Plug the power and duration of each effort into the CP-W' equation and solve for the two parameters. Most modern platforms will fit CP and W' automatically from your power-duration curve if you have hard efforts of varied lengths in your history.
A subtle but important caveat: CP from your existing data is only as good as the efforts in it. If you never go truly maximal for 3 to 5 minutes, the model has nothing to anchor W' with, and your CP estimate becomes unreliable. Real all-out efforts of different lengths are non-negotiable for a trustworthy CP.
Whichever you choose, the same warm-up, terrain, and pacing discipline matter. Garbage efforts produce garbage thresholds, and that is true for both models. Once you have a clean number, the FTP calculator will convert a test result into FTP, watts per kilogram, and a full set of power zones in a few seconds.
What each is good for
FTP is better for standardization and simple structure. If your training revolves around downloadable workouts, prescribed zones, and tracking one trend line over a season, FTP is hard to beat. It plugs directly into Training Stress Score and the rest of the performance-management chart, because TSS is computed relative to FTP. Your easy rides still belong in zone 2 regardless of which model you use, but the zone boundaries themselves are usually drawn as FTP percentages.
Critical Power is better for precision and intermittent efforts. Because CP comes with W', it can model a race full of surges. If you know your W' is 20 kJ and you spend 15 kJ of it covering attacks on a climb, you know you have very little left for the summit. That is information FTP simply cannot give you. CP also separates your aerobic ceiling from your anaerobic capacity, so when you retest you can see which one actually improved.
They answer different diagnostic questions. A flat FTP across a training block looks like no progress. But if your CP held steady while your W' grew, you did improve, just in a different system. The single FTP number hides that. This is one reason riders chasing specific event demands, from criteriums to long gravel, increasingly look at the full power-duration curve rather than one threshold.
The other variable both models lean on is how you read effort within a ride. Average power can hide a workout that was actually a series of hard surges, which is where normalized power earns its place alongside either threshold. A ride can sit below FTP on average while spending real time above CP, draining W' the whole time.
Putting it together: a practical threshold protocol
You do not have to pick a side forever. Most riders are best served by anchoring their day-to-day structure to FTP for compatibility, then layering CP and W' on top for diagnostics and pacing. Here is a workable system.
- Set a baseline FTP with a clean 20-minute test, then take 95% of the average. Use it to define your training zones and to compute TSS.
- Within the same two-week window, ride at least two maximal efforts of different lengths, for example 3 minutes and 12 minutes, on separate days. Let your platform fit CP and W' from those efforts.
- Compare the two thresholds. Note the gap in watts and whether CP reads above or below FTP for you. That offset is yours; remember it.
- Pace long, steady efforts and time trials around CP or FTP, whichever you trust more for hour-plus power. Pace surge-heavy races around CP, and budget your W' deliberately for the climbs and attacks that matter.
- Build intervals off the model that fits the goal. Sweet-spot and threshold blocks key cleanly off FTP. Short, repeated above-CP intervals to grow W' or to extend CP key off the power-duration curve.
- Retest both every 4 to 8 weeks. If FTP stalls, check whether CP or W' moved. Adjust the training emphasis toward whichever system has stopped responding.
- Keep your zones and workout files on one reference so they travel between devices. Mixing FTP-based and CP-based zones in the same plan is how riders end up training the wrong intensities.
The hard part of all this is not the testing, it is the interpretation: holding two related numbers in your head, watching how they move relative to each other over weeks, and turning that into the right session today. That is exactly the synthesis an AI coach is built for. Connect your power data through Strava or your head unit to athletedata.health and your AI cycling coach tracks your full power-duration curve, flags when CP and W' diverge, and prescribes the intervals that target whichever threshold has gone quiet. You can start a 7-day free trial and let it watch the curve for you.
FTP and Critical Power are not rivals. They are two lenses on the same engine, and the riders who improve fastest are the ones who know which lens to look through for the question in front of them.
