It is paradoxical that most plants under natural conditions are infected with vesicular-arbuscular mycorrhizal fungi, yet that it is often difficult to demonstrate that infected plants receive any benefit from the association. The costs and benefits of infection are analysed and a hypothesis formulated that infection only yields benefits at times during the life cycle when P demand by the plant exceeds the capacity of the root system. A simulation model is described that suggests that infection density should be more or less constant below a threshold value of root P uptake rate, but that above this value roots should be non-mycorrhizal. More extensive study of mycorrhizas under field conditions is needed to test such predictions.