Skip to content
Behavioural pattern
Strong research base
Source-checked · Awaiting expert review

The Four Horsemen: how arguments predict breakdown — and what the research really shows

Most people who have argued with a partner can recognise the Four Horsemen from a one-line description. That recognisability is why the idea has spread so far — and why it is worth separating what the research supports from what has been added in the retelling.

Reading time:
8 min read
Written by:
Next Move Editorial · Research-led editorial team; AI-assisted drafting under human direction
Reviewed by:
Expert reviewer not yet appointed
Updated:

The short version

From the 1980s onward, John Gottman, Robert Levenson and colleagues brought couples into a laboratory, asked them to discuss a disagreement, filmed the conversation, measured heart rate and skin conductance, coded every facial expression and remark — and then followed the couples for years to see who stayed together.

Four behaviours during conflict stood out as more common in couples who later separated:

  1. Criticism — attacking the partner's character rather than describing a specific behaviour ("you're so selfish" versus "I felt let down when you forgot").
  2. Contempt — communicating superiority or disgust: sneering, mockery, eye-rolling, sarcasm aimed at the person.
  3. Defensiveness — meeting a complaint with a counter-complaint, an excuse or innocent-victim posture, so that nothing lands.
  4. Stonewalling — withdrawing from the interaction: going silent, looking away, leaving, while physiologically flooded.

What the studies found

Three findings from the longitudinal work are robust enough to rely on.

Negative conflict behaviours predict early separation. Gottman and Levenson's 1992 paper followed couples over four years; those who later dissolved had shown more of the negative behaviours — including defensiveness, stubbornness and withdrawal — during the filmed conflict, alongside higher physiological arousal. Their 2000 follow-up over 14 years found that negative affect during conflict predicted early divorce specifically.

The absence of positive affect predicts later separation. The same 2000 study found that couples who divorced later — often after the first child had reached adolescence — had not necessarily shown more hostility at the start. What they lacked was warmth: humour, affection and interest during both conflict and ordinary conversation. The Four Horsemen are only half the story; the other half is whether anything positive is happening.

How a conversation starts matters. Gottman, Coan, Carrère and Swanson's 1998 study of 130 newlyweds found that harsh start-up — opening with criticism or contempt — and a partner's refusal to accept influence discriminated stable from unstable marriages, while the then-popular "active listening" technique did not.

HorsemanWhat it sounds likeWhy it corrodes
Criticism"You never think about anyone else."Turns a solvable complaint into a verdict on character
ContemptMockery, eye-roll, "oh, poor you."Communicates disgust; the strongest single predictor in Gottman's samples
Defensiveness"Well, you did the same last week."Nothing is heard; the complaint ricochets
StonewallingSilence, turning away, leavingPartner experiences abandonment; conversation cannot repair

The '94% accuracy' claim, and why we do not repeat it

You will often read that Gottman can predict divorce with 90% or 94% accuracy after watching a couple for a few minutes. The figure comes from real papers — the 2000 study reports a 93% classification rate in its sample — but it does not mean what the headline implies.

In 2001 Richard Heyman and Amy Smith Slep published The hazards of predicting divorce without crossvalidation. Their point was methodological: if you build a prediction equation and then test it on the same couples you built it from, it will look far more accurate than it is. When they re-ran a comparable analysis on an independent sample, accuracy fell substantially, and the proportion of couples flagged as "will divorce" who actually did was not meaningfully better than chance.

This is not a refutation of the Four Horsemen. The behaviours really are more common in couples who separate. It is a refutation of the idea that anyone can watch your argument and tell you your future.

Why they escalate

The horsemen tend to arrive together because each invites the next. Criticism provokes defensiveness; repeated defensiveness provokes contempt; contempt provokes flooding, and flooding produces stonewalling — which the critical partner experiences as abandonment, and criticises.

Underneath is physiology. Gottman and Levenson's measurements showed that some partners became physiologically flooded during conflict: heart rate well above resting, adrenaline up, the capacity to take in information sharply reduced. In that state people cannot listen, and most either attack or shut down. This is why the practical advice is not "try harder in the moment" but "change the start, and take a real break when flooded". The regulation skills are set out in Staying steady in a difficult conversation, and the pursue–withdraw loop that stonewalling feeds in Demand–withdraw.

The antidotes

Each horseman has a counterpart behaviour. These are not magic phrases; they are habits that, with practice, change the direction of a conversation.

Criticism → gentle, specific start-up. Describe the situation and your feeling, and make a request. "When the plans changed without a message, I felt sidelined. Could you text me next time?" Three parts: event, feeling, request. No character verdict.

Contempt → appreciation and respect, built outside conflict. Contempt grows in a relationship where the ratio of noticing what is wrong to noticing what is right has tipped. The antidote is deliberately noticing the right: specific, small, said out loud. In the argument itself, the antidote is to stop and name it — "that was contemptuous; I'm sorry" — rather than to defend it.

Defensiveness → accepting some responsibility. Not all of it; some. "You're right that I forgot — I was overloaded, but I did forget." Owning one true piece ends the ricochet and makes it possible for the other person to own theirs.

Stonewalling → announced self-soothing. "I'm flooded. I'm going to take twenty minutes and come back at eight." Then actually calm down — walk, breathe, do something absorbing — rather than rehearsing the argument. The break with a return time is what distinguishes regulation from withdrawal.

And across all four → repair attempts. In Gottman's observations, what separated stable from unstable couples was less the presence of negativity than whether attempts to de-escalate — a joke, an apology, "can we start again?" — were made and accepted. Learning to notice and accept a partner's clumsy repair attempt is one of the highest-value skills in the whole literature.

When it is more than conflict

The Four Horsemen describe conflict between partners with roughly equal power, both of whom can leave the room. Contempt that is constant and one-directional, criticism that polices your friends and movements, or stonewalling used as days-long punishment are closer to patterns of control than to ordinary conflict — and the antidotes above are not the right tool.

Learn next

Frequently asked questions

Research and further reading

Sources marked verified have been checked for title, authorship and publication details.

  1. 1.

    Gottman, J. M., & Levenson, R. W. (1992). Marital processes predictive of later dissolution: Behavior, physiology, and health. Journal of Personality and Social Psychology, 63(2), 221–233 — Observational research on conflict behaviours (including stonewalling and defensiveness) that predict relationship breakdown.

    doi:10.1037/0022-3514.63.2.221 Verified

  2. 2.

    Gottman, J. M., Coan, J., Carrère, S., & Swanson, C. (1998). Predicting marital happiness and stability from newlywed interactions. Journal of Marriage and the Family, 60(1), 5–22 — 130 newlywed couples followed for six years. Criticism, defensiveness, contempt and stonewalling discriminated stable from unstable marriages; accepting influence from a partner predicted stability.

    doi:10.2307/353438 Verified

  3. 3.

    Gottman, J. M., & Levenson, R. W. (2000). The timing of divorce: Predicting when a couple will divorce over a 14-year period. Journal of Marriage and Family, 62(3), 737–745 — Negative affect in conflict predicted early divorce; absence of positive affect predicted later divorce. The widely quoted '93% accuracy' figure comes from this sample and was not cross-validated.

    doi:10.1111/j.1741-3737.2000.00737.x Verified

  4. 4.

    Heyman, R. E., & Slep, A. M. S. (2001). The hazards of predicting divorce without crossvalidation. Journal of Marriage and Family, 63(2), 473–479 — Methodological critique: divorce-prediction equations lose much of their accuracy when applied to a new sample. The reason this site does not quote 'predicts divorce with 90%+ accuracy'.

    doi:10.1111/j.1741-3737.2001.00473.x Verified

  5. 5.

    Christensen, A., & Heavey, C. L. (1990). Gender and social structure in the demand/withdraw pattern of marital conflict. Journal of Personality and Social Psychology, 59(1), 73–81 — Describes the demand–withdraw conflict pattern and shows it depends on who wants change, not only on gender.

    doi:10.1037/0022-3514.59.1.73 Verified

  6. 6.

    Refuge (2026). National Domestic Abuse Helpline (UK): 0808 2000 247, 24 hours. nationaldahelpline.org.uk — Free, confidential UK helpline. Men's Advice Line (Respect): 0808 8010 327.

    www.nationaldahelpline.org.uk/ Verified

  7. 7.

    Ministerio de Igualdad (España) (2026). Servicio 016: atención a víctimas de violencia contra las mujeres (24 h, gratuito, no deja rastro en la factura). violenciagenero.igualdad.gob.es — Spanish national helpline; emergency services on 112.

    violenciagenero.igualdad.gob.es/informacion-3/recursos/telef Verified