REF-750 · Brickley, Doust & Williams, 2002 · European Journal of Applied Physiology, 88, 146–151 · Paper

Ridden at their own critical power, seven trained cyclists lasted between 20 and 41 minutes

Mean about 30 — with heart rate, V̇O₂ and lactate all still climbing at failure: CP is a domain boundary, not a steady state and not a duration

Blimited evidenceMeasured on trained to highly trained athletesWritten before October 2026, being re-checked

Protocol

Seven trained male cyclists (body mass 75.6 ± 6.4 kg, V̇O₂max 4.6 ± 0.7 L·min⁻¹) performed three constant-load rides to failure to derive critical power from the power–duration relationship, then returned on a separate day and rode at exactly that CP until volitional exhaustion, with heart rate, oxygen uptake and blood lactate sampled throughout.

Key Findings

Time to failure at CP ranged from 20 min 1 s to 40 min 37 s; every subject completed 20 minutes and none reached 45. The mean reported in the review literature for this dataset is 29:34 ± 8:22.
Nothing was steady. Over the ride heart rate rose from 118 ± 24 to 177 ± 5 beats·min⁻¹, V̇O₂ from 3.7 ± 0.6 to 4.1 ± 0.5 L·min⁻¹ (approaching each rider's maximum), and blood lactate from 4.3 ± 1.8 to 6.5 ± 2.0 mmol·L⁻¹.
Time to failure correlated with V̇O₂max (r = 0.779, P < 0.05): the bigger engine held its own boundary longer.
The authors' conclusion, verbatim in the abstract: CP "does not represent a sustainable steady-state intensity of exercise".

Read the direction carefully

This is the paper behind a sentence that gets repeated without its source: "CP is sustainable for about 30 minutes". Read what was measured — a mean of 30 with individuals from 20 to 41 — and the sentence loses its precision. Ten minutes either side of the mean is the whole range of what "threshold durability" looks like between riders of similar caliber, and it tracked V̇O₂max, not the CP number itself.

The physiology says why. A CP estimate carries a bandwidth of roughly ±3–5% even with a careful protocol (REF-168): a rider "at" 300 W has an even chance of being just under the boundary, where a steady state eventually arrives, or just over it, where V̇O₂ drifts to maximum and the ride ends. Day-to-day biology moves the true boundary as well. So the same prescribed watts land in different domains on different days, and the clock reads differently each time.

Practical Application

Do not attach a duration to CP. Say what it is — the highest power at which the body can still settle — and, if asked how long, answer "for a trained rider something like 20 to 40 minutes, and yours is a measurement, not a lookup". The same rule applies to MLSS (REF-168) and, a fortiori, to FTP20 (REF-747).
Prescribe "threshold" intervals a few percent below CP unless the point is to visit the severe domain. At exactly CP the athlete is a coin-toss away from a session that cannot reach a steady state; 2–3 × 10–15 min at 95–97% of CP is the heavy domain on purpose, 3 × 8 at 103% is the severe domain on purpose, and "at CP" is neither by design.
Read a fade during a long threshold rep as domain, not weakness. Heart rate and V̇O₂ climbing through a rep held at constant power is what the boundary looks like from the inside; the fix is a lower target or a shorter rep, not a harder athlete.
Track the athlete's own TTE at threshold across a season, on the same protocol, as the durability number it is (REF-747 for the level trend); it moves with V̇O₂max and with training, and it is not comparable between riders.

Limitations

n = 7, men only, one laboratory; the range is the finding and a larger sample would widen it, not narrow it.
CP was derived from three trials on separate days and the TTE ride came after them; ordering and learning effects were not controlled.
Volitional exhaustion at a fixed power has a large day-to-day coefficient of variation; each rider rode one TTE.
Blood lactate and V̇O₂ at end-exercise were still rising, so "steady state not reached" is well supported, but where each rider's true CP sat relative to the estimated one is unknown by construction.