Hill Training for Trail Runners: Train the Half That Decides the Race

Most hill training is uphill training, and that is the half that takes care of itself. Climbing is trainable on a treadmill, a bridge, or the one hill in your town, and it is limited by the same aerobic fitness you already build every week. Descending is the half that decides mountain races, the half that is genuinely hard to train from flat ground, and the half almost nobody schedules.

This guide covers both, weighted the way the race weights them. Our downhill running training sections give you exact, measured doses, not vague advice to "run downhill sometimes." If you are coming from the road and this is your first season on trails, start with your first trail race and come back here when you have a goal race with real vertical.

Hill repeats vs hill sprints vs downhill repeats

Three different trail running hill workouts train three different systems.

The evidence separates them cleanly. Six weeks of uphill intervals improved running economy by 2.4% (Barnes et al. 2013), while two weeks of 30-second uphill sprints raised time to exhaustion with VO2peak unmoved (Kavaliauskas et al. 2018). Four weeks of downhill running raised knee-extensor strength 9.7 to 15.2%, neural drive first and muscle size later, with VO2max unchanged throughout (Bontemps et al. 2022).

Read that last row again. Downhill repeats are the only one of the three that builds eccentric strength in the knee extensors, which is the quality descending actually asks for. Enough downhill volume will train your aerobic system too, the same way any running does, but that is not what you are buying with it: uphill work gives you the same aerobic return far more cheaply, and it gives you no eccentric strength at all.

Uphill work, in brief

Climbing is where trail running is most similar to the running you already do.

Walk the steep parts, and not as a concession. Walking is cheaper than running above about 9.4 degrees, by roughly 8% on average (Giovanelli et al. 2016). Poles buy less energy than people assume, 2 to 3% off the cost of transport, but 14 to 19% off perceived effort (Giovanelli et al. 2019); our poles guide covers where they earn their weight.

Three session archetypes cover it. Long steady climbs for specificity, 1 to 5 minute repeats for aerobic power, short sprints for neuromuscular power. The tested sprint length is about 30 seconds; the shorter 8 to 10 second efforts coaches prescribe on steep ground are practice, not trial evidence. When it comes to form, the focus is the same across the board: lean from the ankles rather than folding at the waist, drive the arms or push on the quads when power hiking, and keep the knee soft.

What climbing does not give you. It does not replace the gym, and it does not prepare your quads for the way down. Six weeks of incline intervals improved economy and lactate thresholds just like flat running did, but left knee strength and muscle power no better than flat or control training did. The authors concluded that uphill running is insufficient as a resistance exercise (Ferley et al. 2014). On the same cohort, flat intervals actually beat uphill intervals for time sustainable at vVO2max (Ferley et al. 2013).

Why descents are the half nobody trains

Going up, your muscles shorten as they produce force. Going down, they produce force while lengthening, and that eccentric braking is what tears muscle fibres. The full mechanism, the damage markers and how to pace descents on race day are covered in our guide to downhill running muscle damage. The short version: the knee extensors absorb most of the negative work, and the strength you lose does not come back between descents.

That fatigue is also what drives the cramps that arrive late in a race, which is one more reason the descents deserve training rather than enduring.

The training consequence is the part that matters here, and it is unusually clean. Descending protects against descending, and climbing does not. In the classic comparison, two 12-minute downhill runs removed the soreness from a later downhill bout, while the group that trained uphill instead got essentially nothing (Schwane et al. 1987). Sports science calls this the repeated bout effect. Runners call it having downhill legs.

Downhill work buys you two different things, and confusing them is why coaches often seem to disagree:

• Protection is cheap and short-lived. Five minutes of easy downhill walking cut the strength loss from a later 40-minute descent by 47 to 64% (Maeo et al. 2017), and the best moment for that light session is 2 to 4 days before the hard effort (Boyd et al. 2023).
• Strength is expensive and slow. It takes weeks of repeated sessions, and it is what the Bontemps and Toyomura numbers describe. Our own Downhill Readiness model is deliberately conservative about how fast this accumulates: a session counts as damaging only above roughly 250 to 300 m (820 to 1,000 ft) of runnable descent, and one hard session reaches only about a third of the maximum protection score.

Two timing rules survive from this distinction, and they are worth keeping whatever your plan looks like. Put the last genuinely hard descent 2 to 3 weeks before race day, because the protection it buys lasts weeks while the fatigue costs days. Put a short easy one 2 to 4 days out, and keep it easy: a light bout protects, another hard one just adds damage.

How to train for downhill running without hills

If you are wondering how to train for vertical gain in flat areas, the answer is that you need a gradient, not a mountain. This is the section where most articles give you a list of five ideas and no numbers, so here are the numbers.

A decline treadmill works, and the evidence is unambiguous. Twenty minutes at 10% decline, three times a week for five weeks, raised knee-extension torque by 9 to 24% and change-of-direction agility by 2 to 3%, with no aerobic gain at that volume. The control group ran flat at matched heart rate and got the mirror image: VO2max up 5%, economy at lactate threshold up 3%, velocity at VO2max up 7% (Toyomura et al. 2018). The two adaptations do not overlap. Runners who dismiss the decline treadmill as useless are wrong about the physiology; what they are right about is availability, since most machines will not go below zero.

Walking down is not a lesser version of running down. One 40-minute session per week of downhill walking at 28% with a pack at 10% of body weight, for four weeks, raised knee-extensor strength at every velocity tested, most of all eccentric torque, up 24% (Maeo et al. 2015). One session a week. No hypertrophy needed. If you have a steep hill and no runnable descent, walking down it repeatedly is real training.

Coaches rank the substitutes (Torrence, iRunFar): race terrain or a camp first, then the steepest local ground, then man-made options, of which a decline treadmill and parking-garage ramps are the practical ones. Runners from flat cities warn that walking back down stadium stairs between reps is its own injury risk, so use the elevator.

Do not turn every descent into intervals. Splitting the same eccentric workload into intervals with rest produced more muscle damage than one continuous bout: strength down 19% versus 13%, twitch force down 31% versus 18%, the deficit still there at 48 hours (Green et al. 2022). Repeats are a convenient format, not a free one.

The recommendations section at the end of this guide puts these doses in order of what they build.

Downhill running technique: Step length, not cadence

If you take one downhill running technique cue into your descent sessions, make it this: shorten the step as the grade steepens, and stop managing cadence. Cadence on its own is a weak lever, and moving far from the frequency you naturally choose costs you muscle activity and energy in either direction. Step length is what changes the load, because on a descent a longer step means the body falls further before the leg catches it, and the steeper the ground the more each extra centimetre costs.

That is worth training deliberately: on downhill sessions pick the step length first and let the legs turn over at whatever rate they want. Our downhill damage guide sets out the measurements, including why raising your cadence does not do what people think.

Match the descents of your goal race

Generic downhill training prepares you for a generic descent. Your race has specific ones, and they are knowable in advance. This is UTMB, with each descent filled by how much eccentric damage it deposits and the worst ones numbered:

The UTMB race page lists every notable descent with its gradient, drop and position on the course. That gives you a target: not "train descents" but "train 8% for 20 minutes at a time, and 15% in shorter pieces, and be able to do it after six hours on your feet".

Then find that gradient near you. Terrain can be indexed by two numbers, how steep and how long, which is exactly the grid the Trail Finder searches:

The Trail Finder holds the descents and climbs pulled from real races and training tracks, mapped by steepness and length, so you can ask for the cell your race lives in and see what exists within driving distance. It is open without an account.

The honest limitation: a slope is a slope in both directions, so the grid does not distinguish a climb from a descent. That works in your favour here, because the 12% climb someone logged is the 12% descent you are looking for.

Surface and technicality: the part you cannot fake

Gradient you can find anywhere. Ground you cannot. Running the identical slope, speed and distance on real trail instead of a treadmill cost 10.5% more oxygen and raised the variability of sideways foot acceleration by 116 to 134%, regardless of how technical the trail was rated beforehand (Nicot et al. 2022). Mild unevenness barely registers, but discrete obstacles push energy cost up 18 to 23% (Bettega et al. 2025).

Here is where we have to be straight with you: no study has tested whether training on smooth ground transfers to technical descending. Not in either direction. The one adjacent data point is discouraging, since runners on an unstable woodchip surface showed no reduction in stride-to-stride variability across five separate exposures (Mohr et al. 2025), though five sessions is a short window.

What experienced trail runners say, consistently, is that smooth-descent competence does not carry to rocks and roots, and that only direct exposure closes the gap. That is practitioner consensus rather than evidence, and it lines up with the measured fact that real ground is a genuinely different task.

The practical version: build the eccentric conditioning wherever you can, then spend your travel budget and your race-week arrival days on the actual surface. If your goal race is technical and you live somewhere smooth, a training camp on that kind of ground is not a luxury, it is the only part of your preparation that cannot be replicated at home.

A word on the last week or two, where experience and evidence pull apart. Runners who arrive early and run easy on race terrain report that the ground stops feeling foreign, and that is worth having. The measurements show no neuromuscular adaptation on that timescale: gait variability on irregular ground did not fall across a 12-minute familiarisation (Sci Rep 2019) or across five exposure days (Mohr et al. 2025). Nobody has tested a one- to two-week block on race terrain against a control, so treat those runs as rehearsal, recon and a pacing check, which is reason enough to do them.

Strength endurance for trail running: work that actually transfers

Strength work does three separate jobs for a trail runner, and they get muddled together constantly. It makes you more economical, it is the main defence against overuse injury, and it is claimed to build endurance under load. The first is well evidenced, the second is well evidenced and covered in the last section of this guide, the third is practice.

For economy, heavy and low-rep wins. Pooled across 22 studies, heavy resistance training beat plyometric training for both running economy and time-trial performance, with near-maximal loads at or above 90% of one-rep max outperforming lighter work, and 10 to 14 week programs outperforming shorter ones (Eihara et al. 2022). Heavy half-squats alone, three times a week for eight weeks, improved economy 5.0% and time to exhaustion 21.3% with no change in VO2max at all (Støren et al. 2008). That is a neuromuscular gain, not an aerobic one.

Descending adds a wrinkle. Downhill 5 km performance was predicted by velocity at VO2max, maximal knee-extensor strength and leg musculotendinous stiffness together, explaining 84% of the variance, while uphill performance did not need stiffness at all (Lemire et al. 2021). Stiffness is exactly what heavy low-rep lifting fails to train directly, which is the case for keeping plyometric and eccentric work in the program. Coaches converge on the same short list: step-downs from a box, Bulgarian split squats with a slow 3 to 4 second lowering phase, single-leg Romanian deadlifts, and box jump-downs.

Worth being clear about what the evidence does and does not say here, because it is easy to read one as the other. No trial has isolated a race-day descending benefit from gym work, and some coaches are openly sceptical that strength training improves ultra performance at all. That is a separate question from injury. A pooled review of 22 trail studies put muscle injuries at 36% and tendinopathy at 24% (Jiang et al. 2024). Strength training is the best-evidenced protection against this overuse, cutting risk by 66% in randomised trials (Lauersen et al. 2018). The tissue it remodels is tendon, increasing its stiffness and cross-sectional area (Bohm et al. 2016). The catch is the clock: strength and neural drive show up in two months, but tendon structural changes take at least three (Kubo et al. 2010). Start strength work early, keep it year-round for injury protection, but get your descending capacity from actual descents.

The other argument is loaded uphill carries for strength endurance. Scott Johnston, who has coached Tom Evans since 2021, calls it muscular endurance work. Preparing for UTMB 2025, Evans climbed the Chamonix vertical kilometre with a 15 kg (33 lb) pack at tempo effort, dropped it at the summit, ran a long fartlek descent and hiked back up at matched heart rate, weekly for seven weeks; his coach describes the same session with roughly 12 kg (26 lb) of water, poured out at the top so the descent is run unloaded. The school's dosing: a grade of 30% or steeper, a load chosen so your legs give out before your breathing does, up to an hour continuous, once a week, in blocks of 8 to 16 weeks.

The recruitment physiology behind it is real, since sustained heavy work pulls in progressively larger motor units as the first ones fatigue. What does not exist is a controlled trial of the protocol against anything, and the nearest evidence is unflattering: vest load degraded trail-running performance progressively at 5 and 10% of body mass (Martinez-Noguera et al. 2024). CTS coaches publicly argue it is unproven and that terrain-specific volume plus conventional lifting does the job better. Our read: for a race with long sustained climbs it is defensible and world-class athletes do it, but do not skip the heavy lifting for it and do not file it under evidence-based.

Key takeaways

• Climbing and descending train different systems. Uphill work is the cheap way to buy aerobic fitness, but six weeks of it left knee strength and power no better than flat running; downhill repeats are the format that builds the eccentric strength descending asks for.
• Muscle damage after a descent is not an injury. Injuries are overuse of tendon and muscle tissue, and strength work is the best-evidenced defence against them, on a timescale of months.
• Muscle protection against descents is the part you can start within days, and one light bout 2 to 4 days out is worth having. Legs that hold up over a long mountain race take weeks of repeated descents.
• You need a gradient, not a mountain. A decline treadmill at 10%, or downhill walking on a steep local slope once a week, produces measurable eccentric strength gains.
• Technical ground is the one thing you cannot simulate. Spend travel on it, and treat race-week runs on it as rehearsal.

Sources: Baggaley et al. 2020 · Vincent et al. 2019 · Martinez-Navarro et al. 2026 · Sheehan and Gottschall 2013 · Wells et al. 2020 · Toyomura et al. 2018 · Bontemps et al. 2022 · Maeo et al. 2015 · Maeo et al. 2017 · Schwane et al. 1987 · Chen et al. 2012 · Nosaka et al. 2001 · Boyd et al. 2023 · Green et al. 2022 · Ferley et al. 2014 · Ferley et al. 2013 · Barnes et al. 2013 · Kavaliauskas et al. 2018 · Giovanelli et al. 2016 · Giovanelli et al. 2019 · Nicot et al. 2022 · Genitrini et al. 2023 · Dhawale and Venkadesan 2023 · Bettega et al. 2025 · Mohr et al. 2025 · Eihara et al. 2022 · Støren et al. 2008 · Lemire et al. 2021 · Martinez-Noguera et al. 2024 · Lauersen et al. 2018 · Bohm et al. 2016 · Kubo et al. 2010 · SAFER SkyRun 2019 · Jiang et al. 2024 · Sci Rep 2019