Estimates of the distribution of the larvae of two species of watermite (Arrenurns spp.) based on values of variance (s2) derived from published accounts of their attachment to their dragonfiy hosts, showed that assumed a non-random, ’clumped’ distribution. The aggregation of infective stages within the range of the host, which is thus refiected, is considered as indicating that the association with the host is primarily dispersive, as in phoresy, rather than parasitic. The requirements and mechanisms for limiting attachment frequency are discussed. Variations in the degree of ’dumping’ with increasing mite load suggest that searching is random but that host discovery is greatest when the mite is most numerous.