Shared cM Relationship Predictor
Shared cM relationship predictor. Enter the total centimorgans your DNA test reports for a match and see every relationship consistent with it, ranked by probability, with the expected range for each and an honest warning when several fit equally well. Runs entirely in your browser.
Shared cM Relationship Predictor
Every relationship that fits
The numbers behind the ranking
| Relationship | Probability | Mean cM | 5th–95th | z | Share nothing |
|---|
Run the simulation yourself
How to use the shared cM predictor
Find the total centimorgans
Open the match on AncestryDNA, 23andMe, MyHeritage or FamilyTreeDNA and read the total shared cM. Use the total, not a percentage and not the length of the longest segment.
Enter it
Type the figure or drag the slider. The shortcut buttons jump to the values that mark the boundaries between relationship groups, which is a quick way to see how sharply the answer changes.
Read the whole list, not just the top
The bars show every relationship consistent with your figure. If the tool warns that the result is ambiguous, that is the finding: several relationships genuinely fit and the number alone cannot separate them.
Check the spread, then check the simulation
The table gives each candidate’s mean, its 5th-to-95th-percentile range, how many standard deviations your figure sits from its mean, and how often that relationship shares nothing at all. The panel below re-runs the underlying model live so you can watch the spread appear.
Reading a centimorgan figure honestly
A DNA test hands you a number and an implied certainty that the number does not carry. Your match shares 1,750 centimorgans; the site suggests “close family”. But 1,750 cM is the expected amount for a grandparent, for a half sibling, and for an aunt or uncle — three quite different people. The arithmetic cannot choose between them, and a tool that names one is not being more useful, it is being less honest.
This predictor takes the opposite approach. It ranks every relationship consistent with the figure you enter, shows the expected range for each, and says plainly when the top candidate is not meaningfully ahead of the next two. Where the answer is genuinely sharp — a total near 3,400 cM is a parent or a child and nothing else — it says that too.
Where the numbers come from
The distributions are not borrowed. Every relationship here is built as a real pedigree and simulated: chromosomes are laid out at their published genetic lengths, crossovers are placed along them by the standard two-pathway model that reproduces the observed spacing between crossovers, genomes descend through the pedigree, and the shared amount is measured. Twenty thousand replicates per relationship give the mean, the spread and the probability of sharing nothing at all.
Simulating rather than copying was a licensing decision as much as a technical one. The compilation everyone quotes carries a licence we could not settle to our own satisfaction, so nothing from it is embedded here. Its published averages are used the way published figures should be used — as external truth to check our own model against. On that check the model does well: the simulated means track the published averages for near relationships, and the probability that third and fourth cousins share nothing comes out at 7% and 49% against 8% and 52% in an independent study of 100,000 simulated pairs.
Absence of shared DNA is not absence of ancestry. About half of all genuine fourth cousins share nothing a test can see.
Why the spread matters more than the average
Only one relationship in the whole subject has no variation: a parent and child are half-identical across the entire genome, every time. Everything else is a lottery of which chunks got passed down. Full siblings average about 2,530 cM but any given pair lands somewhere in a band hundreds of centimorgans wide, and by the third and fourth cousin range the distribution has spread so far that it overlaps two or three neighbouring relationships completely.
That is why the table reports a 5th-to-95th-percentile range next to every mean, and a z score telling you how many standard deviations your figure sits from each candidate. A match two standard deviations from a relationship’s mean is not impossible, only unusual — and unusual things happen constantly across a match list of thousands.
A note for readers in Southeast Asia
Consumer DNA testing has grown quickly across Singapore, Malaysia and the wider region, and it arrives into family structures that the tests were not designed around. Populations with a long history of marriage within a community share more DNA than the same relationship would elsewhere, which pushes matches to look one step closer than they are. If your family comes from such a community, treat every prediction here as an upper bound on closeness rather than a reading, and lean harder on documents and shared matches.
10 facts about shared DNA
A parent and child share every centimorgan — the one relationship with no variation at all.
Full siblings average about three quarters of a genome copy, not half.
A grandparent, a half sibling and an aunt all sit near 1,700 cM.
Roughly half of all fourth cousins share no detectable DNA whatsoever.
Segments under 7 cM are discarded by most consumer tests.
The autosomal genetic map used here totals about 3,391 cM.
Women produce roughly 1.6× as many crossovers per meiosis as men.
Crossovers are not independent — interference spaces them out along a chromosome.
A single cM figure is consistent with several different relationships.
Everything here is computed in your browser; nothing is uploaded.
Frequently asked questions
-
A centimorgan is a unit of genetic distance, not physical length: it is the stretch of chromosome across which about one crossover happens per generation. DNA testing companies report a match as the total centimorgans two people share, because that total maps far more directly onto how closely related they are than a count of matching letters would. The autosomal map used by this tool totals about 3,391 cM.
-
Because several relationships genuinely have the same expected amount. A grandparent, a half sibling and an aunt or uncle all sit near 1,700 cM, and no amount of arithmetic on that one number can separate them. This tool says so plainly and shows the whole ranked list rather than picking a winner. Genealogists resolve the ambiguity with ages, documents and shared matches.
-
They are simulated here, from published chromosome map lengths and a published model of how crossovers are placed along a chromosome. Every relationship is built as a real pedigree, genomes are dropped through it 20,000 times, and the resulting spread is measured. Nothing is copied from a third-party dataset; the simulated averages are then checked against the figures published in the literature.
-
The simulated means match the published averages closely for near relationships — full siblings, half siblings, aunts, first cousins — and the probability of sharing nothing at all comes out at 7% for third cousins and 49% for fourth, against 8% and 52% in an independently published study of 100,000 simulated pairs. The predictions are as good as that model, which is to say good enough to shortlist relationships and never good enough to prove one.
-
Because the biological answer is fixed but the reported one is not. A child is half-identical to a parent across the entire genome, so the true variance is zero; every centimorgan of spread you see between tests is measurement — different maps, different segment thresholds, different genotyping. The predictor allows for that explicitly, which is why a 3,450 cM result is still confidently a parent or child.
-
It is the probability that two people genuinely related in that way share no detectable DNA at all. It is essentially zero for anything closer than a second cousin, rises to about 7% for third cousins, and reaches roughly half for fourth cousins. This is why a DNA test cannot disprove a distant paper relationship: absence of shared DNA is not absence of ancestry.
-
Yes. Half siblings, half aunts and uncles, half first cousins, and once- and twice-removed cousins are all in the candidate list, which matters because they are exactly the relationships that collide with the full ones. Half first cousins and first cousins once removed, for instance, share an expected amount within a few centimorgans of each other.
-
Any of them, as long as it is a total in centimorgans rather than a percentage. Totals differ a little between companies because they use different maps and discard segments below different lengths, and this tool models a 7 cM floor. Treat differences of a few percent between providers as normal rather than as a contradiction.
-
No, and any tool that claims otherwise is overstating its case. Shared DNA narrows the possibilities, sometimes to one or two, and combines with everything else you know. It is evidence, weighed alongside ages, records and other matches — which is precisely how professional genealogists treat it.
-
Completely free, with no account and no limit. The whole calculation, including the live simulation, runs in your browser, so the centimorgan figure you enter never leaves your device and the page keeps working offline once it has loaded.
Pick up where you left off
Stored only in this browser — never sent to our servers.