France 2027Presidential polls and forecast

France 2027 Election

Wins out of 100 simulated elections

How to read this chart

What you see

  • One square = one simulated election, out of 100, coloured by the camp that won it.
  • Solid: the camp’s main candidate.
  • Hatched: another candidate of the same camp, when they are the one running (Bardella instead of Le Pen, for instance).

Where the number comes from

  1. Who runs? Each simulation first draws the candidates, using probabilities published with their reasons. An absent candidate’s votes go to the candidates closest to them.
  2. Starting point: today’s polling average, with its own margin.
  3. Historical correction: the front-runners are pulled towards the field, as in past campaigns (10 of 11 front-runners fell back).
  4. Uncertainty: what can still move before the vote, plus an error shared by all polls, with surprises like in 2002.
  5. Two rounds: the top two qualify, then the runoff is simulated.
  6. Count: 20,000 simulations, scaled to 100 squares.

Today

RN: 80 wins out of 100 simulated elections, with 172 days to go before the first round.

  1. RN80 Le Pen 69Bardella 11
  2. Centrist bloc10 Philippe 9Attal 1
  3. Place publique-PS6 Glucksmann 6
  4. Mainstream right3 Retailleau 2Lisnard 1
  5. LFI1 Mélenchon 1

Assumptions by candidate

How to read this table

What you see

  • Three nested bars, from the palest to the darkest: runs, makes the runoff, wins. Each bar sits inside the one before.
  • %A camp row: its total wins, the same as in the grid.

Where the number comes from

  1. Runs: an editorial estimate based on past elections, not a poll. Each probability is explained in the methodology.
  2. Makes runoff: the share of simulations in which the candidate finishes in the top two. A candidate who does not run cannot make it.
  3. Wins: the share of simulations in which they win the runoff.

Today

Marine Le Pen: 85% runs 76% makes the runoff 69% wins.

Runs Makes runoff Wins
RN 80%
Marine Le Pen RN 85% 76% 69%
Jordan Bardella RN 15% 13% 11%
LFI 1%
Jean-Luc Mélenchon LFI 95% 40% 1%
Centrist bloc 10%
Édouard Philippe Horizons 80% 26% 9%
Gabriel Attal Renaissance 55% 5% 1%
Place publique-PS 6%
Raphaël Glucksmann Place publique 90% 26% 6%
Mainstream right 3%
Bruno Retailleau LR 60% 6% 2%
David Lisnard Nouvelle Énergie 45% 4% 1%
Reconquête <1%
Éric Zemmour Reconquête 75% 2% <1%
Other candidates 45–90% <1% <1%

How the forecast has changed

How to read this chart

What you see

  • One line per camp: its chance of winning, week by week.
  • A note: a campaign event. Hover for the date and the source.
  • A dotted vertical line: the simulated lineup changed that day.

Where the number comes from

  1. The line is replayed. For each past week, the whole calculation is run again with only the polls published by then.
  2. Same rules as today: same lineup, same candidacy probabilities. A move in the line comes from the polls, never from a change to the model.
  3. A camp = the sum of its candidates’ chances. A candidate not yet polled at a date is left out of that point.
  • RN
  • Centrist bloc
  • Place publique-PS
  • Mainstream right
  • LFI
  • Debout!

First-round polling average

How to read this chart

What you see

  • The line: the polling average, candidate by candidate.
  • The light band: an 80% chance that the true average of the day lies inside. It is not the election-day forecast, which is much wider (see the scenarios).
  • One dot = one poll, larger and darker when it weighs more in the average.

Where the number comes from

  1. Weigh: a poll counts according to how recent it is, its sample size (capped at 3,000) and the pollster’s rating. Polls from the last 14 days carry 88% of the weight.
  2. Correct: a pollster that always gives a candidate a little more (house effect) is brought back towards the others.
  3. Translate: a poll testing a different lineup is brought to the most probable one; absent candidates’ votes go to the politically closest.
  4. Smooth: a Kalman filter treats opinion as something that moves a little each day, and polls as noisy measurements of it. It works on the ratios between candidates, so the total is always 100%.

Today

Le Pen stands at 35.5% in the average, from 75 polls and 319 tested lineups.

  • Le Pen · RN
  • Mélenchon · LFI
  • Philippe · Horizons
  • Glucksmann · Place publique
  • Retailleau · LR
  • Zemmour · Reconquête
  • Tondelier · Les Écologistes
  • Dupont-Aignan · Debout la France
  • Roussel · PCF
  • Arthaud · LO

Candidate scenarios

How to read this chart

What you see

  • A|BEach button sets a lineup: the most probable one by default, then variants.
  • The light bar: the first-round share in 80% of the simulations.
  • The dark bar: the middle half of the simulations.
  • The tick: the median, the “middle” share.
  • The hollow dot: today’s polling average, for comparison.

Where the number comes from

  1. Why do the bars sit away from the dots? In the 1995–2022 presidential elections, 10 of the 11 candidates polling at 20% or more five months out fell back. The model therefore pulls the front-runners towards the field. With 172 days to go, a front-runner keeps about 64% of their lead over the field; the effect fades out by election day.
  2. Why so wide? Two sources of chance, calibrated on history: what can move during the campaign (6.4 points for a 20% candidate), and the polling error on election day (1.7 points). Their distribution makes big surprises more frequent than a normal one would.
  3. Scenario or main forecast? A scenario does not draw the candidacies: it shows what would happen if this lineup were the one on the ballot. It is simulated 20,000 times with the same draws as the main forecast: a difference between two scenarios comes from the lineup alone.

Today

Le Pen 35.5% in the polls 27.2% on average in the simulations, between 17 and 38% in 80% of them.

The bars show each candidate’s election-day vote share across the simulations; the dot shows today’s polling average. The simulations pull candidates towards each other: in past campaigns, 10 of the 11 candidates polling at 20% or more five months out ended up lower.

CandidateMakes runoffWin
Le Pen88%80%
Philippe34%10%
Glucksmann26%6%
Retailleau9%3%
Mélenchon40%<1%
Zemmour3%<1%
Tondelier1%<1%
Dupont-Aignan<1%<1%
Roussel<1%<1%
Arthaud<1%<1%
Averages and ranges
CandidateAverage80% range95% rangeElection day
(mean)		Election day
(80%)		Makes the
runoff		Tops the
first round
Le Pen RN35.532.9–38.031.7–39.527.217.3–37.788%65%
Mélenchon LFI17.015.5–18.714.7–19.616.78.6–25.740%12%
Philippe Horizons15.714.2–17.213.5–18.115.87.9–24.834%10%
Glucksmann Place publique13.512.0–14.911.4–15.714.36.6–23.226%8%
Retailleau LR7.36.5–8.26.0–8.79.73.7–16.69%3%
Zemmour Reconquête4.23.7–4.83.5–5.16.22.2–10.93%1%
Tondelier Les Écologistes2.42.1–2.81.9–3.03.61.2–6.41%<1%
Dupont-Aignan Debout la France1.91.7–2.21.5–2.42.81.0–5.1<1%<1%
Roussel PCF1.81.6–2.11.4–2.32.70.9–4.9<1%<1%
Arthaud LO0.70.6–0.80.5–0.91.00.3–1.9<1%<1%

Possible runoffs

How to read this chart

What you see

  • %The percentage on the right: the share of simulations in which this runoff happens.
  • The split bar: each finalist’s expected share if this runoff happens. The bracket: where the split falls in 80% of the simulations. The tick in the middle: 50%.
  • The 100-square grid: who wins this runoff, out of 100 simulations in which it happens.

Where the number comes from

  1. Who qualifies? In each simulation, the top two of the first round. Each runoff’s frequency follows.
  2. Two sources, blended by precision: the polls of this runoff, if it was tested at least three times, and a vote-transfer model: voters of eliminated candidates split between the finalists by political proximity, some abstain. This model misses polled runoffs by about 4.5 points. A runoff polled often mostly follows its polls; one never polled rests on transfers alone.
  3. The first round matters: a finalist who beats expectations in the first round has more reserves. A runoff-specific error and drift, calibrated on the 1995–2022 runoffs, are added.
  1. Le Penvs.Mélenchon 33% chance of this runoff
    63.9%between 58 and 69% 36.1%between 31 and 42%
  2. Le Penvs.Philippe 27% of simulations
    54.2%between 48 and 60% 45.8%between 40 and 52%
  3. Le Penvs.Glucksmann 20% of simulations
    55.1%between 49 and 61% 44.9%between 39 and 51%

Other runoffs: 20% of simulations

Runoff polls

How to read this chart

What you see

  • 1·2·3One panel per runoff listed above, in the same order, with the same candidate named first.
  • The line and its band: the share of the candidate named first in the panel’s title (the line in their colour), in the runoff polls of the last 12 months, and the 80% range of that average.
  • The dots: the polls. The horizontal line: 50%; above it, that candidate wins.
  • Empty axes: this runoff was never polled over the period.

Where the number comes from

  1. Same method as the first round: polls weighted by recency, sample size and the pollster’s rating, smoothed by a Kalman filter.
  2. The polls alone. The runoff forecast above adds the vote-transfer model and everything that can move before the vote, so the two figures can differ.

Le Pen – Mélenchon

21 polls · latest Oct 22

Le Pen – Philippe

22 polls · latest Oct 22

Le Pen – Glucksmann

11 polls · latest Oct 22

Latest polls

How to read this table

What you see

  • %The figures published by the pollsters, as they are, in % of valid votes. No adjustment is applied in this table.
  • BEach row: fieldwork, pollster and its rating, sponsor, sample, method, link to the filing with the Commission des sondages.
  • ↓“Show all polls” opens the full table, its filters and the Weight column: each poll’s exact share of today’s average.

Where the number comes from

  1. Recency: a recent poll weighs more. Those from the last 14 days carry 88% of the average.
  2. Sample: more respondents, more weight, up to 3,000.
  3. Pollster rating: weight multiplied by 0.8 to 1.2.
  4. No double counting: waves from the same pollster less than 10 days apart share their weight, and the lineups tested in one poll count as one observation.
  5. Margin of error of a 1,000-person poll: from ±1.4 points for a 5% candidate to ±3.1 points at 50%.
Pollfieldwork · pollster · sponsor · sample Le PenMélenchonPhilippeGlucksmannRetailleauLisnardAttalZemmourTondelierRuffinDupont-AignanRousselArthaud Weight
Oct 21–22 OpinionWay B CNews · 973 resp. · online · notice ↗ 3616121366441—110.5 12.7%
Oct 21–22 Ipsos B+ Le Parisien · 1,500 resp. · online · notice ↗ 321714135.56.5—43.5—1.52.50.5 4.5%
Oct 19–22 Cluster17 B 2,817 resp. · online · notice ↗ 3318.516138——42—320.5 4.5%
Oct 19–21 Toluna Harris Interactive B+ 2,094 resp. · online · notice ↗ 351713126—562—211 3.2%
Oct 17–19 Ifop B− 1,928 resp. · online · notice ↗ 341512125.55.5432.51.51.520 1.7%
Oct 14–16 Elabe A− 1,701 resp. · online · notice ↗ 31.515.515.5156.55.5—32.5—31.50.5 1.5%
Oct 12–15 YouGov C+ Le HuffPost · 1,094 resp. · online · notice ↗ 3513812546452220 1.0%
Oct 14–15 Cluster17 B Le Point · 1,263 resp. · online · notice ↗ 3118171177—32—21.50.5 0.8%
Oct 13–14 OpinionWay B CNews · 1,043 resp. · online · notice ↗ 3515121364552—111 2.8%
Oct 13–14 OpinionWay B 1,001 resp. · online · notice ↗ 351412137—743—230.5 0.8%
Show all polls (75)

Pollster ratings

How to read this table

What you see

  • AThe rating, from A+ to C. B is the average pollster; a dash means too short a track record.
  • −0.3Relative error, in points: the pollster’s past error minus the expected one; negative when it missed by less than its competitors in the same elections, so better. The rating divides it by the average poll’s own error (see the rating scale under the table).
  • +0.4Bias left, centre, right, far right: the pollster’s past miss for each bloc, relative to the other pollsters; positive when it overstated the bloc.
  • ×1.1Weight: the multiplier applied to its polls in the average.

Where the number comes from

  1. A track record: 541 polls published in the 60 days before past elections (presidential since 1995, European since 2009, legislative since 2017), compared with the official results.
  2. A fair comparison: each error is compared with the other pollsters’ in the same election at the same time. In 2022 they all underestimated Mélenchon: the rating keeps only what sets a pollster apart.
  3. Caution: with few polls, the rating is pulled towards the average; under 5 polls, no rating and a slightly lower weight.
  4. Effect: the rating sets the weight, from ×0.8 to ×1.2. The bloc bias is published but does not adjust the 2027 polls: methods change from one election to the next.
Pollster Rating Mean error
(pts)		 Relative
error		 Bias
left		 Bias
centre		 Bias
right		 Bias
far right		 Elections
rated		 Polls
rated		 2027
polls		 Weight
Ipsos B+A− to B− 2.03 −0.03 +0.1−0.1+0.3−0.3 13804 1.06
Toluna Harris Interactive B+A to B− 2.05 −0.05 −0.2−0.1−0.2+0.4 96016 1.09
LH2 B−B to C 2.15 +0.07 +0.4+0.2−0.4−0.2 3160 0.89
Verian B+A to B− 2.22 −0.05 +0.1−0.10.00.0 9532 1.09
Odoxa A−A+ to B+ 2.25 −0.08 −0.2+0.3−0.3+0.1 5165 1.16
YouGov C+ 2.27 +0.14 −0.4−0.3−0.5+0.7 452 0.79
Elabe A−A+ to B+ 2.32 −0.09 −0.4+0.20.0−0.2 7346 1.18
CSA C+B− to C 2.37 +0.12 +0.1+0.2−0.7+0.5 5330 0.82
Ifop B−B to C+ 2.38 +0.06 0.00.0+0.2+0.3 1210121 0.91
BVA B−B+ to C+ 2.39 +0.04 0.0−0.1−0.1−0.1 11490 0.94
Cluster17 BA to C+ 2.39 −0.02 +0.6−0.3−0.1−0.8 4287 1.04
OpinionWay BA− to B− 2.58 −0.01 −0.1+0.1+0.6−0.2 106112 1.01
AtlasIntel — Limited track record: default weight 110 0.83
Le Terrain — Limited track record: default weight 120 0.83
Viavoice — Limited track record: default weight 220 0.83
  • AExpected error at least 7% below the average pollster’s (A+, A, A−)
  • BBetween 7% below and 10% above the average (B+, B, B−)
  • CMore than 10% above the average (C+, C)
  • —Limited track record: default weight

Methodology

Polls

This site tracks every published voting-intention poll for the 2027 presidential election since March 2023: 75 first-round polls, covering 319 candidate lineups, and 31 runoff polls as of October 22, 2026. The list comes from Wikipedia and is checked against the filings pollsters must make with France’s polling regulator, the Commission des sondages. Each poll is shown with its pollster, sponsor, fieldwork dates, sample size and method.

French pollsters often test several lineups of candidates in the same poll. Each lineup is mapped onto a reference field: the declared candidates that pollsters test regularly. When a candidate is missing from a lineup, we estimate where their voters go from polls that tested both lineups on the same respondents.

Average

The average weights each poll by how recent it is, its sample size and its pollster’s rating. It is smoothed over time, and the band around each line shows the uncertainty in today’s average.

Pollsters are rated on their track record across 13 elections: the presidential elections from 1995 to 2022, the European elections from 2009 to 2024 and the legislative elections from 2017 to 2024. The gaps between them are small: weights range from 0.8 to 1.2 and barely move the average.

Simulations

The model runs the election 20,000 times. Each simulation starts from the current average, with the front-runners pulled back towards the pack: in past campaigns, 10 of the 11 candidates polling at 20% or more five months out ended up lower. It then adds two sources of uncertainty: how opinion may shift before the vote, and the error all pollsters share on election day. Both are calibrated on the presidential elections from 1995 to 2022, for major and minor candidates alike. The top two in each simulation go through to the runoff, which is decided by the head-to-head polls or, for a matchup that has never been polled, by estimated vote transfers.

A candidate’s chance of winning is the share of simulations they win.

Candidacies

The field is not official yet, and the headline figure accounts for that. Each simulation first draws who is on the ballot, based on each candidate’s chance of running, and never keeps two candidates from the same camp when only one of them can run, such as Marine Le Pen and Jordan Bardella. Results are then grouped by camp.

These chances are editorial estimates based on past elections. Since 1995, six months before the first round, 96% of the most-polled candidates made it onto the ballot, as did 75% of the minor candidates polled regularly and 30% of the figures polled only occasionally.

The scenarios further down each assume a fixed field, the most likely one by default, and first-round polling is mapped onto that field. Each candidate’s chance of running, and the reasoning behind it, is published with the data.

Track record

In past presidential elections, the model’s 80% range contained the result 79% of the time five months before the vote, 84% three months out, 81% one month out and 84% a week out. These rates rest on six elections and about thirty candidates at each point, so a gap of 10 points from the target is within the range of chance. In 2017 and 2022, runoff polls overstated the far right by 3 to 4 points. The model does not adjust for this, as there are too few elections to go on; if it happened again, Marine Le Pen’s chance of winning would be a few points lower.

Updates

The forecast is updated every night. Under France’s law of 19 July 1977, no poll is published or discussed on this site on the day before or the day of each round, and the forecast is frozen during that time. The simulations shown here are calculations based on published polls. Every poll has a margin of error.

Forecast sensitivity
If…Le PenBardellaPhilippeGlucksmannRetailleau
Baseline assumptions69%11%9%6%2%
Campaign drift halved74%12%6%5%1%
Campaign drift multiplied by 1.566%11%10%8%2%
No drift (opinion stays put until election day)78%13%5%3%<1%
Systematic polling error multiplied by 1.569%11%9%7%2%
Normal tails instead of Student’s t72%11%8%6%1%
0.5 correlation between candidates of the same bloc69%11%9%6%1%
Runoff error and drift multiplied by 1.564%10%10%8%2%
First-round shocks carried into the runoff, not capped at its calibrated drift60%9%11%9%3%
Runoff polls ignored (vote-transfer model only)62%11%12%10%2%
Most likely field, no uncertainty over who runs80%<1%10%6%3%
Every declared, regularly polled candidate runs74%<1%10%6%3%
Twice the risk that candidates drop out63%26%5%5%<1%
Half the risk that candidates drop out72%5%10%6%2%
Election-day shares centred on today’s average, no uncertainty about reversion to the mean79%14%3%2%<1%
Centred on today’s average, reversion to the mean drawn in either direction (raw shares)79%14%3%3%<1%
Shares revert towards the field average (log scale), rate estimated without shrinkage69%11%8%7%1%
Shares revert towards the field average (raw shares)75%13%5%4%1%
If the 2017 and 2022 runoff polling miss recurs (far right overstated in the runoff, average error at this point in the race)65%10%11%8%2%