How Many People Marched? Why Police and Organizers Count Differently

On Saturday, October 24, 2026, French authorities estimated that 410,000 people marched nationwide, including 110,000 in Paris. The CGT labor union put the figure at about 800,000 nationwide and 250,000 in Paris. Reports noted that crowd estimates varied sharply. Which is right? Statistics cannot settle that, but it can show why the numbers differ and how to handle them.

The size of the gap

The organizers' figure is about 1.95 times the authorities' nationwide (800,000 ÷ 410,000) and about 2.3 times in Paris (250,000 ÷ 110,000). The nationwide gap is 390,000 people, nearly as large as the authorities' own estimate. The Paris share of the total also differs: 27% under the authorities' figures (110,000 ÷ 410,000) and 31% under the organizers'. The two sides disagree not only on the total but on its geography.

Why counts diverge

Three things drive such gaps. Definitions: a march lasts hours and people join and leave, so counting the peak crowd gives a smaller number than counting everyone who took part at any time. Methods: estimates can come from area and density, from counting people passing a fixed point, or from reports gathered by phone. Incentives: each side may prefer the number that supports its message. None of this requires bad faith. It only means the figures answer slightly different questions.

Area times density

The simplest method multiplies the area filled by the density of people. Suppose, hypothetically, that a march fills a street 2 km long and 25 m wide, an area of 50,000 square meters. At a loose density of 1 person per square meter that is 50,000 people; at a dense 3 per square meter it is 150,000. A threefold range comes from one assumption alone, before counting anyone who passed earlier or arrived later. The scientific calculator handles the arithmetic, and the exercise shows how sensitive a headline number is to a guess about density.

Counting a sample

A better method divides the route into equal cells, counts people in a random sample of cells and extrapolates. Suppose 10 randomly chosen cells, out of 400, contain 12, 18, 15, 22, 9, 14, 20, 17, 11 and 16 people (invented counts). The descriptive statistics calculator gives a mean of 15.4 and a standard deviation of about 4.1. The confidence interval calculator for a mean gives a 95% interval of about 12.5 to 18.3 people per cell. Multiplying by 400 cells yields an estimate of about 6,160, with a plausible range of roughly 5,000 to 7,300. Even a careful sample leaves a margin of about ±19%, and it works only if the cells are chosen at random.

A single crowd number is an estimate. An honest report gives a range and a method.

What to do with two numbers

Do not average the two estimates. A figure of 605,000 would look precise but would have no more evidence behind it than either source. It is more accurate to report both, name who produced each, and say that the true figure is uncertain. Treat any ratio built on a crowd size with the same care. Interior Minister Laurent Nuñez said 135 people had been arrested by midday on October 24, when about 35,000 had gathered, roughly 0.4%. That was a snapshot, not a final rate, and the denominator was itself an estimate.

Questions to ask

Who produced the number, and how? Is it a peak count or a total? Is the figure for the whole country or one city? Has it been revised? Is a range given? These questions work for any rally, strike or demonstration, in any country. This article takes no position on the protests or on the government's response.