John Gottman's laboratory at the University of Washington was an apartment with cameras in it. Couples came in and were asked to discuss an area of ongoing disagreement for fifteen minutes while their interaction was recorded and their physiology monitored. The tapes were then coded second by second, expression by expression, using a system derived from Ekman's action units.
Gottman's headline claim, repeated widely, is that this coding predicted which couples would divorce with something around ninety percent accuracy. That number deserves care: several of the published prediction rates come from models fitted to the same sample they predict, which inflates accuracy, and independent prospective testing has produced weaker results. The statistical criticism is fair and worth knowing.
What has held up better is the descriptive finding about which behaviors are corrosive, and one of them stands out from the rest.
Gottman's Four Horsemen are criticism, defensiveness, stonewalling and contempt. Contempt is the strongest predictor, and it is qualitatively different from the other three.
It is coded as a specific facial action: a unilateral tightening and raising of one lip corner, often with the head tilted back so the other person is looked at from above, frequently with an eye-roll. Ekman's taxonomy places it near disgust, and the phenomenology matches — contempt is disgust directed at a person rather than a substance, with an added dimension of hierarchy.
That hierarchical component is what makes it corrosive in a way that anger is not. Anger, however unpleasant, is an engagement between equals: I am furious with you about this. Contempt communicates something structurally different — that the other person is beneath the level at which their view requires consideration. And because that is the message, it blocks the mechanism by which relationships repair themselves. Gottman's work on repair attempts found that what distinguishes stable couples is not the absence of conflict but the success of small bids to de-escalate. Contempt makes those bids fail, because a bid from someone you hold in contempt is not received as a bid.
The physiological findings are consistent with the severity: recipients of contempt show elevated stress markers, and Gottman reported associations with immune function measures.
The behavior generalizes well outside marriage, which is why it is in this guide. A manager who displays contempt toward a subordinate has already decided the relationship's structure, and no amount of process will fix it. A political exchange conducted in contempt is not a disagreement about policy; it is a claim about who counts. A negotiation in which one side displays contempt has revealed that they are not negotiating.
As an observable, it is unusually good, because it is fast, specific, and hard to fake in the absence of the underlying attitude — that unilateral action is not one most people can produce on demand. And it is often the only channel through which a settled judgment leaks, because contempt is precisely the attitude people are most careful to keep out of their words.
If you see it in a relationship that matters to you, it is not a mood. It is a report on the current hierarchy, and it arrived before anyone said anything.
See Defense: detecting contempt before collapse