When to put the heating on, from 76 years of daily readings
Autumn turns cold about ten days later than it did in the 1990s, and six weeks apart north to south. Why the honest answer is not a date at all.
Somebody in your house wants the heating on. Somebody else says it is too early. Neither of you has any evidence, and the argument will be settled by whoever is colder.
This is a question a calendar cannot answer, and most of the advice that gets published in September pretends otherwise. So here is what this site’s own record says instead.
What “turns cold” means here
This site stores the daily maximum, minimum and mean temperature for twenty five British towns, every day from 1950 to the present. That is enough to ask a question no almanac answers: in a given town, when does autumn actually turn.
The definition used below is deliberately simple. For each year, take the first day from 1 August onwards on which the average of that day and the six days before it falls below 12C. One cold afternoon does not count. A week that has genuinely cooled does.
The 12C threshold is this site’s own choice, not anybody’s standard, and the reason for it is worth stating. Twelve degrees sits several degrees below the summer peak of even the coolest town in the set, so the crossing marks a real turn in the season. Pick a threshold closer to the summer peak and the northern towns cross in early August, which tells you about the threshold rather than about the weather.
Two honesty notes. These are reanalysis readings, each an average over a grid cell of about eleven kilometres, so they run a shade smoother than the hottest or coldest station in the same town. And 2026 is left out of every figure below, because it is not a complete year.
Ten days later than the 1990s
Averaged across all twenty five towns, the crossing point fell on 25 September through the 1990s. Across the six complete years from 2020 to 2025 it fell on 5 October. That is about ten days later.
Every single one of the twenty five towns crossed later in the 2020s than it did in the 1990s. Not most of them. All of them. For a set of towns running from Plymouth to Inverness, with no two of them sharing a climate, that unanimity is the most interesting number in this piece.
It is not a smooth trend, and pretending otherwise would be dishonest
Average crossing date across the twenty five towns, by decade.
| Decade | Average crossing date | Complete years |
|---|---|---|
| 1950s | 25 September | 10 |
| 1960s | 23 September | 10 |
| 1970s | 21 September | 10 |
| 1980s | 24 September | 10 |
| 1990s | 25 September | 10 |
| 2000s | 5 October | 10 |
| 2010s | 28 September | 10 |
| 2020s | 5 October | 6 |
The 2000s were already as late as the 2020s, and the 2010s fell back a week. Fitted as a straight line across all 76 complete years, the crossing moves later by about 1.6 days per decade. That is a real signal and a modest one, sitting inside a lot of year to year noise, and anyone quoting the 1990s to 2020s gap as a rate of change is overstating it.
Six weeks separate Inverness from Plymouth
This is the part that makes national advice useless.
| Town | 1990s average | 2020s average | Days later |
|---|---|---|---|
| Plymouth | 16 October | 31 October | 14 |
| Brighton | 15 October | 29 October | 14 |
| Portsmouth | 15 October | 28 October | 14 |
| Southampton | 15 October | 26 October | 12 |
| Norwich | 1 October | 13 October | 12 |
| Cardiff | 4 October | 13 October | 9 |
| London | 3 October | 11 October | 9 |
| Cambridge | 2 October | 9 October | 7 |
| Bristol | 1 October | 9 October | 8 |
| Reading | 29 September | 7 October | 8 |
| Oxford | 28 September | 3 October | 4 |
| Liverpool | 28 September | 2 October | 4 |
| Leicester | 28 September | 1 October | 2 |
| Nottingham | 28 September | 30 September | 2 |
| Sheffield | 22 September | 29 September | 7 |
| Birmingham | 21 September | 29 September | 9 |
| Leeds | 19 September | 29 September | 11 |
| Manchester | 17 September | 29 September | 11 |
| Belfast | 14 September | 28 September | 14 |
| Newcastle upon Tyne | 15 September | 27 September | 12 |
| Edinburgh | 6 September | 22 September | 15 |
| Glasgow | 8 September | 21 September | 13 |
| Perth | 9 September | 17 September | 8 |
| Dundee | 5 September | 16 September | 10 |
| Inverness | 1 September | 15 September | 14 |
About six weeks separate the top of that table from the bottom. A household in Plymouth and a household in Inverness are not having the same argument in the same week, and they never were. Your own town, and the rest of the stored record behind these figures, is at heat data.
Your own year will not match your own average
The averages above are the least useful part of the table for deciding what to do this week, and it is worth being blunt about why.
Take London across the six years of this decade. The crossing landed on 30 September in 2022, 4 October in 2024, 8 October in 2020, 15 October in 2021, 18 October in 2023 and 23 October in 2025. That is a spread of more than three weeks in six years. Across all twenty five towns and all 76 years, the typical year lands about thirteen days away from its town’s own long run average, in one direction or the other.
So the record can tell you that autumn is arriving later than it used to, and that your grandmother’s rule about the first of October was already wrong for most of the country when she made it. What it cannot do is tell you about this October. Nothing can.
Watch the house, not the calendar
Which leaves the only measurement that actually decides the question, which is the temperature in the room you are sitting in.
The NHS advises heating your home to a temperature that is comfortable for you, and, if you can, at least 18C in the rooms you regularly use such as the living room and the bedroom. It notes that this matters particularly if you have a health condition. That is a floor, not a target, and it is a floor that applies in the middle of September as much as in the middle of January.
If a room you use is sitting below it, the heating question is answered, whatever the date. If it is comfortably above it and you are reaching for the thermostat out of habit, the date is answering a question the house has not asked. A cheap indoor thermometer in the coldest room you regularly use is worth more than every seasonal heating article published this month, including this one.
Two groups of people should not wait for the argument to be settled by whoever is colder. Older people, and anyone with a heart or lung condition, lose heat faster and feel it later, and a cold room does measurable harm rather than just being unpleasant. If somebody in the house falls into either group, the thermometer decides and not the calendar.
What October costs, and why the answer changed this year
The other half of the decision is the bill, and it moved on 1 October 2026.
Gas is capped at 7.97p per kWh for a typical direct debit customer, up from 7.33p, while VAT came off domestic electricity. That narrowed the gap between the two fuels to its smallest in years and changed which machine is cheaper to heat with. The arithmetic, the verified rates and the crossover point are in what the October 2026 price cap changes about what you run.
If what you are heating with is electric, a plug-in heater or a reversible air-to-air heat pump run in heat mode, the running cost calculator will give you the cost per hour from the rated power and the rate on your own bill. It has no default rate on purpose.
What actually moves the date for your house
Three weeks of difference between two houses in the same street is ordinary, and none of it is about the weather.
The building fabric is most of it. A house that holds heat needs the boiler on later and for less time, and the same fabric that keeps winter warmth in is what keeps summer heat out. That is one conversation running in both directions, not two, and what it costs to make an old house cope with heat is the summer half of it with the itemised bill attached.
The reverse case is worth knowing too, because it catches people out in September. A heavy British house that has been absorbing warmth all summer gives it back for weeks, which is exactly why a top floor bedroom is still hot at midnight in August. The same thermal mass is why the ground floor of that house can feel cold first thing in October while the bedroom is still fine, and why heating the whole house because one room is chilly is usually the wrong move.
And if the reason you are asking is a cold snap rather than the season turning, the live warnings page carries the Met Office severe weather warnings, cold as well as heat, for wherever you are.
The answer to the question at the top, then, is that there is no date. Autumn turns about ten days later than it did in the 1990s, six weeks apart from one end of the country to the other, and a fortnight either side of that in any given year. Put a thermometer in the coldest room you use, keep it above 18C when somebody is in there, and let the house tell you.
Questions
- Is there a right date to put the heating on?
- No, and this site's own record is the reason. Across twenty five British towns the point at which autumn turns cold is spread over about six weeks, from the middle of September in Inverness to the end of October in Plymouth, and within a single town it moves by roughly a fortnight from one year to the next. Any national date is wrong for most of the country most of the time.
- What indoor temperature should I be heating to?
- The NHS advises heating your home to a temperature that is comfortable for you, and if you can, at least 18C in the rooms you regularly use, such as the living room and the bedroom. It says this matters particularly if you have a health condition. Treat 18C as a floor rather than a target.
- Has autumn really got later, or does it just feel that way?
- It has moved, but less smoothly than the phrase suggests. On this site's definition the average crossing point across twenty five towns fell on 25 September through the 1990s and on 5 October across 2020 to 2025, about ten days later. Fitted across all 76 complete years the trend is about 1.6 days per decade, and the 2000s were already as late as the 2020s, so this is a shift with a lot of noise in it rather than a clean ramp.
- Which town moved most?
- Every one of the twenty five moved later, which is the finding worth more than any single town. The largest gaps between the 1990s and the 2020s, of a fortnight or more, were in Edinburgh, Plymouth, Brighton, Portsmouth, Belfast and Inverness. The smallest were two days, in Leicester and Nottingham.
- Where do these dates come from?
- This site stores the daily maximum, minimum and mean for twenty five British towns back to 1950. For each year it takes the first day from 1 August onwards on which the average of that day and the six before it falls below 12C. The threshold is this site's own choice, picked because it sits several degrees below every town's summer peak, so the crossing is a real turn in the season rather than noise near a ceiling.
The figures in this piece were computed by this site from daily weather readings it stores and processes. Underlying readings from the Open-Meteo historical archive, licensed CC BY 4.0.