Scope. This page provides a controlled reader-run protocol and does not claim site-run measurements. For cooking, cooling, storage and reheating, follow the current official guidance linked on the page.
Recipes give one temperature for "fish." Fish is not an ingredient. It is thousands of species with different muscle structures, different fat contents, different connective tissue and different collagen behaviour, and the temperature that makes one of them perfect makes another of them chalky.
This page is built to replace that single number with a per-species reference: the temperature to pull at, the window either side of it where the result is still good, the point where it stops being good, and the visual cue that corresponds to each. It is the flagship reference for everything in this section and it feeds the Temperature Ladder.
It is also a page about a food where safety questions and texture questions get tangled together constantly, usually to the detriment of both. We keep them apart, deliberately and visibly, and the reason is explained below.
Two questions, two sections, kept apart on purpose
There are two entirely different questions people ask about cooking fish.
"What temperature makes this taste right?" That is a texture question. It is answered by measurement: a probe, a log, replicates and a stated failure point, taken on the fish and the equipment in front of you.
"What temperature makes this safe?" That is a food-safety question. It is answered by named public authorities, it varies by jurisdiction, by species and by the population being cooked for, and it is not ours to derive, round, simplify or interpolate.
These two answers are frequently different numbers, and merging them produces the worst kind of advice: a single figure that is presented as authoritative on a question it was never measuring. So this page has a doneness section and a safety section, they are separated by a rule and a heading, and neither one should ever be read as speaking for the other. If you take a texture number from the table below and treat it as a safety clearance, you have misread the page, and we would rather say that plainly than trust the layout to communicate it.
The safety section links each point to the named authority's own page, and nothing in the doneness section overrides it.
Part one: doneness (texture)
Temperatures below are internal temperatures at the thickest part of the fillet, measured at the moment of pulling from the heat unless the row says otherwise. Record Celsius as canonical, with Fahrenheit alongside it if you want both.
The table
Every row carries provenance. A filled row is either measured, with a log reference, or cited, with a named source and a retrieval date. Until then it is a reader check and its cells stay blank. Do not estimate a species from a similar one, and do not carry a number over from a recipe.
| Species | Pull temperature | Window | Failure point | Visual cue at target | Provenance |
|---|---|---|---|---|---|
| Cod | — | — | — | — | Reader check |
| Haddock | — | — | — | — | Reader check |
| Hake | — | — | — | — | Reader check |
| Plaice | — | — | — | — | Reader check |
| Sole | — | — | — | — | Reader check |
| Turbot | — | — | — | — | Reader check |
| Sea bass | — | — | — | — | Reader check |
| Sea bream | — | — | — | — | Reader check |
| Salmon | — | — | — | — | Reader check |
| Trout | — | — | — | — | Reader check |
| Mackerel | — | — | — | — | Reader check |
| Sardine | — | — | — | — | Reader check |
| Tuna | — | — | — | — | Reader check |
| Swordfish | — | — | — | — | Reader check |
| Monkfish | — | — | — | — | Reader check |
These rows are reader-check fields, not a promised site benchmark. Fill one only from a dated run on the stated species and geometry, with its log attached; otherwise leave it blank rather than borrowing a number or guessing.
Why one number for "fish" cannot work
Three properties vary enormously across species, and each of them moves the window.
Fat content. Fat conducts heat differently from lean muscle and, more importantly, it changes how much you notice moisture loss. An oily fish that has lost some water still eats as moist. A very lean white fish that has lost the same proportion eats as dry. The same absolute overshoot is a small error in one and a ruined dinner in the other.
Muscle structure. Fish muscle is arranged in short blocks separated by thin sheets of connective tissue, which is why cooked fish flakes rather than shredding. How thick those sheets are, and how much collagen is in them, differs by species and decides both the flaking behaviour and how the fish holds together at the pull temperature.
Collagen behaviour. Fish generally has less connective tissue than the braising cuts used elsewhere on this site. Because this article has not established a dependable cross-species temperature comparison, it makes no precise collagen-temperature claim.
Add geometry to those three. A thick loin and a thin tail-end fillet of the same species behave differently because the heat has different distances to travel, so the gradient between the surface and the centre is different, and the amount of carryover after pulling is different too. A single per-species number, from anyone, is always an approximation over shape.
Lean white fish
The least forgiving category, and the one where the difference between right and ruined is smallest.
With little fat to mask moisture loss, the transition from translucent and yielding to opaque and chalky is quick and it is not reversible. These are the species where pulling early and letting carryover finish the job matters most, and where gentle methods pay the largest dividend. If you want a wider margin rather than better reflexes, this is the category to cook in oil or in a low oven rather than in a hot pan.
Visual cue at the target, described rather than numbered: the flesh has just turned from glassy to opaque at the surface, still shows a slightly translucent core at the thickest point, and separates into flakes under light pressure without falling apart. We give no probe reading for that cue; take one yourself the next time you see it and write it into the table.
Oily fish
More forgiving, for the reason above, but with its own failure mode. Oily fish that is taken too far does not go chalky in the same way. It goes dry and the flavour changes character as the fat is driven out and oxidises, which is a different and less obvious failure.
The useful consequence is that the window here is wider and the failure is later, so these species are the sensible place to learn to use a probe. We give no target or window for them; find yours with the probe and the table above.
Some species in this group are commonly served with a deliberately underdone centre. That is a texture preference on the doneness axis and a separate question on the safety axis, and we are not going to blur them here. See the safety section.
Firm and steak-cut fish
Species with dense muscle and steak-like cuts behave differently again. They tolerate handling, they hold together under methods that would break a delicate fillet, and several of them have culinary traditions of being served well below the doneness normal for white fish.
Two practical notes that do not require a number. First, thickness dominates in this group, so timing advice transfers even more poorly than usual and probing matters even more. Second, carryover in a thick steak is larger than in a thin fillet simply because there is more hot mass, so the gap between the pull temperature and the peak temperature is bigger. We give no figure for that gap; measure it by leaving the probe in after the pull and noting the peak.
How to probe a fillet without ruining it
A doneness table is worth nothing if the probe is in the wrong place, and fish is the hardest common food to probe well. This section matters as much as the numbers.
- Go in from the side, along the length, rather than down through the top. A side entry lets you place the tip in the middle of the thickness without leaving an obvious hole in the presentation surface.
- Find the thickest point first by eye and by touch, and probe there. The thickest point is the last to reach temperature, and it is the one the window applies to.
- Watch for the skin and the pan. A probe tip that has gone too far and is touching the pan, the tray or the skin will read the equipment rather than the fish, which is the most common way home readings go wrong.
- Use a thin probe. A thick sensor in a thin fillet damages a meaningful share of the food and takes longer to settle.
- Read the movement, not the first number. Insert, watch the reading climb, and take the value where it stops rising. A single glance at a still-climbing probe under-reads.
- Probe once, decide once. Every entry is a channel for moisture to leave. This is not a food to check five times.
- Pull before the target, because it will keep rising off the heat. How much it rises is the subject of what happens ninety seconds past the window.
For very thin fillets, a probe may not be usable at all and the honest answer is that the visual cue is the instrument. Where that is the case, note it on the row rather than recording a number nobody can act on.
Part two: safety (cited authorities only)
This section answers a different question from everything above. Nothing here is derived, measured or adjusted by us. Every statement is reproduced from a named authority with its retrieval date, or it is absent. Where authorities disagree, we publish the disagreement rather than choosing.
Each point links to the authority's own page. This section signposts; it does not summarise.
Cooked-fish guidance. FDA consumer guidance says most seafood should reach 145°F (63°C). That safety endpoint is separate from any texture target in the table above. Guidance checked 21 August 2026.
Fish served raw, cured or lightly cooked. Gravlax, ceviche, tartare and deliberately rare preparations sit outside cooked-fish guidance. FDA explains that some fish can contain parasites, prior freezing can address parasites, freezing does not kill every harmful germ, and thorough cooking remains the safest route on its consumer seafood-safety page.
Parasite treatment. Do not improvise a freezer schedule. Commercial and retail controls are specified in the FDA Food Code, §3-402.11, with species exceptions and record requirements; whether domestic equipment can verify those conditions is a separate question this page does not answer.
Vulnerable groups. FDA identifies pregnant people, children, older adults and people with weakened immune systems as higher-risk groups and advises them to avoid raw or undercooked fish and shellfish on its consumer seafood-safety page. Use your own authority's current advice directly.
Storage, chilling and holding. FDA says seafood intended for use within two days should be refrigerated at 40°F (4°C) or below; otherwise freeze it. See the full handling guidance, not a texture target, for safety decisions.
We take no position on sourcing ethics or sustainability anywhere on this page. That is a real subject and it is not this page's job.
What this page does not cover
Publishing the gap is part of the method:
- Every species row above is reader check. No pull temperature, window or failure point has been measured.
- Carryover by thickness is outside this page's scope for any species, so the pull-versus-peak gap is unknown.
- Method comparison across a low oven, oil poaching, water poaching and a covered pan is outside this page's scope for the same species and thickness.
- The visual cues have not been paired to readings. They are descriptions, not correlations; pairing them with your own probe readings is the most useful thing you can add to the table.
- Shellfish are out of scope on this page and get their own reference, because their windows are narrower still.
- Frozen-and-thawed versus never-frozen is outside this page's comparison. Do not transfer a texture result between those conditions without a matched test.
The admission rule is permanent: a populated row must carry its log reference. A changed number belongs in the corrections record rather than being quietly edited.
Where To Go Next
- What happens ninety seconds past the window, which is the mechanism behind the failure point column.
- Oil poaching, the most forgiving method, if you would rather widen the margin than sharpen your timing.
- How to cook a whole fish, where probing and geometry get considerably harder.
- How To Judge A Piece Of Fish In Ten Seconds, for the stage before any of this.
- Shellfish: The Narrowest Windows In The Kitchen, for the category this page excludes.
- Gravlax And Cold Cures, Safely, where the method and the cited guidance travel together.
In Plain English
What it means. There is no single correct temperature for fish, because fish is not one material. Fat content, muscle structure and thickness all move the window, and the window is narrow enough that the difference between good and ruined is small. Separately, and it is genuinely separate, there is safety guidance that comes from public authorities rather than from anybody's kitchen.
Who it helps. Anyone who has bought one expensive piece of fish and has a single attempt at it, and anyone who has followed a recipe's timing exactly and produced something dry.
When to use it. At the hob, with a probe in your hand, deciding whether to pull now or give it another thirty seconds. That is the moment this page is built for.
What to do next. Use the structural parts of this page, which need no numbers: probe from the side at the thickest point, pull before you think you should, and treat lean white fish as having far less margin than oily fish. For safety questions, go to the authority, not to us.
Take the chart to the hob
The doneness window chart is free, ungated and designed to be printed and stuck somewhere splashable. It carries both scales, a provenance column and blank rows for you to fill from your own runs. Print the doneness window chart and fill it in as you cook.