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.
Custard has a reputation for being temperamental, which is a way of saying that nobody has told you the numbers. It is not temperamental. It is narrow. There is a temperature at which it begins to thicken, a range across which it thickens usefully, and a temperature above which it stops being custard and becomes sweet scrambled eggs in warm milk. That last transition is not gradual and it is not reversible.
Published in degrees, the whole thing becomes a manageable engineering problem with a small margin. Published as "cook until it coats the back of a spoon", it becomes a lottery you will sometimes lose without ever knowing why.
Recording the window
This is what the page is for, so it goes first rather than after eight hundred words of context.
Worksheet fields (stirred egg custard, no starch): recipe and batch size; pan and heat setting; the temperature at which it first thickens; the range across which it thickens usefully; the failure temperature, where it first turns grainy; and the carryover rise after it leaves the heat, with pan and volume recorded. Copy these labels into a notebook or spreadsheet and fill one row per run.
Two of those rows do not appear in recipes at all. The failure temperature tells you how much room you have, and the carryover figure explains why so many custards curdle after the pan is off the heat. This page supplies neither as a site result: measure them with a calibrated probe and keep the log attached, or cite a named, dated source.
This article does not reproduce isolated-protein set points as though they were a custard recipe. The window that matters is the one measured on the whole mixture, with ingredients and method recorded.
What is physically happening on either side of it
Below the window, the yolk proteins are still folded and dispersed. The mixture is warm milk with egg in it, and it will stay that way indefinitely at that temperature. Time does not substitute for temperature here. Holding a custard base below the set point for an hour will not thicken it.
Inside the window, the proteins unfold and begin to link to one another, forming a loose three-dimensional network. Liquid is trapped inside that network, and the mixture thickens. Critically, the network at this stage is fine enough to be invisible: the custard is smooth because the protein strands are too small and too evenly distributed to feel.
Above the window, the same links keep forming and the network keeps tightening. Strands aggregate into visible clumps, and the clumps squeeze out the liquid they were holding. What you see is a grainy, separated mixture with watery liquid around solid specks. That is the same mechanism as an overcooked scrambled egg, which is why the same eggs at eight temperatures is the best possible primer for this page: a custard is the same protein doing the same thing, diluted in dairy and sugar, which slows it down and buys you a slightly wider window than a bare egg gets.
The dilution matters and it is worth being explicit about why. Sugar and dairy get physically in the way of the proteins finding each other, so both the set point and the failure point shift upwards relative to a plain egg. The size of that shift depends on the sugar and dairy quantities, which is one reason a window quoted from one recipe does not automatically transfer to another.
The failure temperature, and how fast you cross it
The window's width is the whole story, and the width is small enough that ordinary rates of heating can carry you across it faster than you can react.
Consider what a pan is doing. The heat is applied to the base, the base is hotter than the liquid, and the liquid closest to the base is hotter than the bulk. A probe in the middle of the pan reads the bulk, not the hottest part. By the time the bulk reading enters the top of the window, the layer at the base has been past the failure temperature for some time, and the specks you eventually see were formed down there minutes before the bulk caught up.
This has three consequences that recipes rarely draw out:
- Slow the heat input, not just the target. The gap between base temperature and bulk temperature is the thing that ruins custards, and lowering the heat narrows that gap. A water bath narrows it to nearly nothing, at the cost of time.
- Stir the base and the corners, not the middle. The stagnant film against the metal is where the damage starts. A flat-edged implement that actually reaches the corner of the pan does more than vigorous stirring in the centre.
- The pan does not stop when the hob does. Residual heat in a heavy base will push the custard further after removal, and a custard sitting in a hot pan can cross the failure temperature with nobody touching it. Have the sieve and the cold receiving bowl ready before you start, and decant immediately rather than "letting it settle". This is a common way for a technically correct custard to be lost, and it happens after the cook thinks they have finished.
What starch does to the window
This is the most useful thing on the page, and it is missing from almost every custard recipe that is not explicitly a pastry cream recipe.
A custard thickened by egg alone and a custard containing starch are not the same dish with a small variation. They have different failure behaviour, and the difference is large.
Worksheet fields: make an egg-only custard and a starch-containing one side by side, and record for each the temperature at which it thickens, whether it survives being brought to the boil, and its texture after a night in the fridge. Copy these labels into a notebook or spreadsheet and fill one row per run.
For this article, the defensible claim is operational: starch-thickened custard and unstarchened custard are different systems. Compare their measured behaviour directly; do not use a molecular story or a claimed enzyme threshold to fill a gap in the record.
The practical upshot, which you can act on today:
- If your custard recipe contains flour or cornflour, boiling it is not a mistake, and stopping short of a boil is a common cause of a pastry cream that thins in the fridge.
- If your custard is egg and dairy only, the boil is fatal.
- Adding a small amount of starch to an egg-only custard is a legitimate insurance policy against curdling, at the cost of a slightly different texture and a less clean flavour. Whether that trade is worth it is a matter of taste, and we are not going to pretend it is a rule.
Judging it without a probe, and why you should use one anyway
The traditional test is that the custard coats the back of a spoon and holds a line when you draw a finger through it. It works, in the sense that it is genuinely correlated with the custard having thickened. The problem is that it is a lagging indicator. The spoon test tells you that the protein network has already formed enough to change the viscosity you can see, which means the reading was reached some seconds earlier and the base of the pan reached it earlier still.
Every visual cue in custard-making has this property. By the time the change is visible in the bulk, the leading edge of the reaction is somewhere ahead of it. A probe reads the bulk too, but it reads a number rather than an impression, and it reads it continuously, so you can see the rate of climb and pull before it becomes a problem. Rate is the information you actually need, and no visual cue gives you rate.
If you have no probe: keep the heat low enough that the climb is slow, stir the base constantly, and pull at the first sign of thickening rather than the confident sign. The difference between "I think something is happening" and "that is definitely coating the spoon" is often most of your remaining margin.
If you have a probe: calibrate it, find your own window with the worksheet above, and pull to a number. This is the entire reason the site exists.
If it has already curdled
Honest hierarchy, best case first:
- Caught within seconds, faintly grainy. Off the heat immediately, into a cold bowl, and whisk hard. Some of the aggregation is fine enough to be broken up mechanically. This works often enough to be worth trying and not often enough to rely on.
- Visibly grainy but not separated. A stick blender, then straining through a fine sieve. You are breaking the clumps small enough to be imperceptible rather than reversing anything. The texture will be slightly heavier than it should have been.
- Separated, with liquid around solids. It is gone. Protein aggregation is not reversible, and no amount of blending puts the water back inside the network. Do not add more heat, do not add more egg, and do not try to cook it into submission.
What to do with a lost custard rather than binning it: strain it, keep the liquid, and use it as the dairy component of something that does not depend on texture. The flavour is intact. It is only the structure that has failed.
Treat both rescue routes as reader checks rather than guaranteed fixes. On a small batch, record the temperature at failure, the rescue attempted, and whether the custard becomes smooth and stays smooth after cooling. Discard the attempt if food-safety guidance for its ingredients or time out of refrigeration has not been followed.
If your problem is a repeated failure rather than a single one, nine things that kill a custard is organised by symptom, and the tempering step in particular is worth reading on its own terms in how to temper eggs properly.
A note on raw eggs
Some custard preparations are cooked to a temperature well below anything that constitutes cooking in a food-safety sense, and some are not cooked at all. We do not improvise guidance on that. Egg-safety guidance, with the named authorities linked, is in raw-egg preparations. Nothing on this page should be read as a safety statement.
In Plain English
A custard thickened by egg alone has a narrow working window and a hard failure point above it, and crossing that point is irreversible. Most curdling happens either at the base of the pan, where the liquid is hotter than the probe reads, or after the heat is off, when residual pan heat keeps pushing. Adding starch widens the window considerably and changes the rules entirely: a starch custard can and usually should be boiled, while an egg-only custard cannot.
Who this helps: anyone whose custard works most of the time and fails occasionally, without a pattern they can name.
When to use it: crème anglaise, ice cream base, pouring custard, quiche filling, anything set with egg alone. And every time you read a recipe that says "do not let it boil" without telling you what the actual limit is.
What to do next: decide which kind of custard you are making before you turn the heat on, because it determines whether the boil is a failure or a requirement. Set up the sieve and the cold bowl first. Put a probe in the pan, keep the heat low, stir the base rather than the middle, and pull to a number instead of to an impression.
Next Step
The same pattern, a working range with a hard failure point above it, runs through every emulsion in how to find every classic emulsion's working range and break point. How to read the Temperature Ladder explains how set points on this site are labelled measured, cited or reader check.