This paper asks whether the gap in subjective happiness between spouses matters per se, i.e. whether it predicts divorce. We use three panel databases to explore this question. Controlling for the level of life satisfaction of spouses, we find that a higher satisfaction gap, even in the first year of marriage, increases the likelihood of a future separation. We interpret this as the effect of comparisons of well-being between spouses, i.e. aversion to unequal sharing of well-being inside couples. To our knowledge, this effect has never been taken into account by existing economic models of the household. The relation between happiness gaps and divorce may be due to the fact that couples which are unable to transfer utility are more at risk than others. It may also be the case that assortative mating in terms of happiness baseline-level reduces the risk of separation. However, we show that assortative mating is not the end of the story. First, our results hold in fixed-effects estimates that take away the effect of the initial quality of the match between spouses: fixed-effects estimates suggest that a widening of the happiness gap over time raises the risk of separation. Second, we uncover an asymmetry in the effect of happiness gaps: couples are more likely to break-up when the difference in life satisfaction is unfavourable to the wife. The information available in the Australian survey reveals that divorces are indeed predominantly initiated by women, and importantly, by women who are unhappier than their husband. Hence, happiness gaps seem to matter to spouses, not only because they reflect a mismatch in terms of baseline happiness, but because they matter as such.