If I’m autistic, will my child be autistic too?

What the family studies found

The largest single piece of evidence comes from a Swedish national cohort of 2,049,973 children born between 1982 and 2006, of whom 14,516 were diagnosed with autism.

It measured relative recurrence: the rate of autism in someone with an affected family member against the rate in someone without one. For full siblings the figure was 10.3, with a confidence interval of 9.4 to 11.3. In absolute terms that’s 829 cases per 100,000 person-years among those with an affected sibling, against 49 among those without (Sandin et al., 2014).

The gradient across relationships is the part worth noticing, because it tracks genetic closeness almost exactly:

Relationship	Relative recurrence
Identical twin	153.0
Full sibling	10.3
Fraternal twin	8.2
Maternal half sibling	3.3
Paternal half sibling	2.9
Cousin	2.0

A meta-analysis of twin studies reaches a compatible conclusion by a different route. Pooling seven studies that met its recruitment criterion, it found a correlation of .98 between identical twins and .53 between fraternal twins when autism prevalence was set at 5% (Tick et al., 2016).

Identical twins share effectively all their DNA; fraternal twins share about half, and both kinds of pair usually share a household. A large gap between the two correlations is what points to genetic influence over shared-environmental influence, which is why twin designs carry so much weight here. The same analysis also ran an extremes analysis on autism trait scores in a population sample of 6,413 twin pairs, extending the finding beyond diagnosed cases into the trait distribution.

That second point connects this page to a broader pattern: autistic traits are continuously distributed through the population, and the genetic influences on the diagnosis appear to overlap with those on milder trait variation.

Why a heritability figure isn’t a personal probability

This is where the question usually goes wrong, and the distinction matters more than any number on this page.

Heritability is a population statistic. It estimates what proportion of the variation in a characteristic across a group is associated with genetic differences within that group. It says nothing about what any individual carries or transmits, and it changes if the population or the environment changes.

A concrete way to see the gap: the Swedish study’s figure of 10.3 for full siblings is a ratio between two rates, both of which are low in absolute terms. Eight hundred and twenty-nine per 100,000 person-years is a real elevation over 49. It’s also not a majority of anything.

Ratios are easy to misread in exactly this direction. “Ten times higher” sounds like a probability when it’s a comparison between two small numbers, and the absolute figures are the ones that describe what happens to most families in that cohort.

So “autism is highly heritable” and “your child will probably be autistic” are different claims, and only the first has evidence behind it. Nothing in this literature produces a percentage for one prospective parent.

Where the estimates disagree

The figures are not as settled as summaries suggest, and the disagreement is worth seeing.

The same Swedish cohort that produced the recurrence gradient estimated heritability at 0.50, with a confidence interval of 0.45 to 0.56, and reported that the data supported a model of additive genetic and non-shared environmental effects (Sandin et al., 2014). The twin meta-analysis landed considerably higher.

Both are competent studies. They differ because heritability estimates depend on study design, on how broadly autism is defined, and on what population rate is assumed—the meta-analysis itself reports different correlations depending on the prevalence used.

The practical upshot is that anyone quoting a single confident percentage is flattening a real spread. What survives the disagreement is the direction: a substantial genetic contribution, consistent across methods.

It’s worth knowing which numbers tend to circulate. The higher twin-study figures are the ones that reach popular summaries, because they’re more striking. The population-cohort estimate of around 50% is drawn from two million children and national registers, which is a different and in some respects stronger kind of evidence. Neither is wrong; they’re answering the same question with different instruments in different samples.

What the data doesn’t cover

Three limits belong on this page.

These studies measured siblings and twins, not parents and children. That’s the recurrence data that exists at this scale. Family clustering is well established, but a precise parent-to-child figure isn’t what these particular studies produced, and it would be a misreading to apply the sibling number to your own child.

Diagnosis rates aren’t fixed. Recognition has broadened considerably since the Swedish cohort was born, particularly for adults and for women. A family where one person is diagnosed is also a family more likely to notice traits in others, which affects measured recurrence without anything genetic changing. That ascertainment effect pushes recurrence figures upward for reasons that have nothing to do with inheritance, and it’s difficult to separate out.

Autism isn’t one thing genetically. The research describes many contributing variants of small effect instead of a single inherited factor, which is part of why prediction at the individual level isn’t available. There’s no test, and there’s no counselling protocol that produces a number for an autistic parent.

Nothing here predicts severity or support needs. Recurrence and heritability concern whether autism is present, not what form it takes. Two autistic people in one family can need different kinds of support.

The environmental share isn’t zero. The Swedish study’s model included non-shared environmental effects alongside additive genetic ones, and the twin meta-analysis was written partly in response to debate about how much shared environment contributes. Anyone presenting autism as purely genetic is overstating a literature that continues to argue about the split.

What the question is often asking

The literal question has a narrow answer. The question underneath it usually doesn’t.

Some people are asking whether they’d be passing on difficulty. That framing deserves examining before answering: what makes autism hard is a mixture of the thing itself and how well the surrounding world fits, and the second half is not fixed. An autistic parent tends to be unusually well placed to recognise traits early and to build a home that accommodates them, which is a real advantage that no recurrence figure captures.

Some are asking whether to have children at all. That’s a values question, not an evidence question, and it isn’t one a page or a statistic should answer. It’s worth saying plainly that the research on this page has historically been used to argue positions its authors weren’t making, and none of it constitutes an argument for or against anyone having a family.

Some are asking how to talk to a partner about it. The useful version of that conversation is about what you’d both want to do if a child turned out to be autistic—what you’d notice early, what you’d push for, what you’d refuse—which is a more answerable question than a probability nobody has.

And some are asking what to watch for. That’s answerable: know what early traits look like, take your own observations seriously, and expect to be better than average at noticing. Autistic parents frequently report recognising things in a child that professionals initially dismissed, which is worth trusting.

One framing worth avoiding entirely is fault. Heritability describes a statistical association across a population; it doesn’t make any parent responsible for a child’s neurotype, and reading it that way imports a judgement the data doesn’t contain. A child being autistic isn’t an outcome that needed preventing.

If you’re weighing your own recognition and not a child’s, what happens in an adult autism assessment covers the process and our guide to autism in women covers the presentation most often missed. Where families are concerned, the useful preparation is knowing what you’d want to do differently from how you were raised.