Many birds and mammals store energy as hoarded food supplies. A supply of stored food is beneficial during periods when food is scarce, but building up and managing such a supply also entails costs. The optimal number of caches will be reached when the net benefit is at its maximum. If dominants can steal more stored food from subordinates than the other way around, the optimum will differ between these categories. A previous theoretical model of hoarding in groups with dominant and subordinate members produced three testable predictions: (1) hoarders should store more food as anticipated future conditions get worse; (2) subordinate flock members should store more food than dominants; and (3) dominants should increase hoarding relatively more than subordinates as conditions get worse. Here we present a field experiment on willow tits (Parus montanus) designed to test these predictions. We found support for all three. Hoarding increased as conditions got worse, subordinates stored at a higher rate than dominants, and dominants increased their hoarding effort relatively more than subordinates as conditions worsened. These results support the incorporation of information on dominance and food availability into models predicting food storage behaviour.