The matrilineal Mosuo of southwest China live in large communal houses where brothers and sisters of three generations live together, and adult males walk to visit their wives only at night; hence males do not reside with their own offspring. This duolocal residence with ‘walking’ or ‘visiting’ marriage is described in only a handful of matrilineal peasant societies. Benefits to women of living with matrilineal kin, who cooperate with child-care, are clear. But why any kinship system can evolve where males invest more in their sister's offspring than their own is a puzzle for evolutionary anthropologists. Here, we present a new hypothesis for a matrilineal bias in male investment. We argue that, when household resources are communal, relatedness to the whole household matters more than relatedness to individual offspring. We use an inclusive fitness model to show that the more sisters (and other closely related females) co-reside, the more effort males should spend working on their sister's farm and less on their wife's farm. The model shows that paternity uncertainty may be a cause of lower overall work rates in males, but it is not likely to be the cause of a matrilineal bias. The bias in work effort towards working on their natal farm, and thus the duolocal residence and ‘visiting marriage’ system, can be understood as maximizing inclusive fitness in circumstances where female kin breed communally.