Virtually all higher organisms form holobionts with associated microbiota. To understand the biology of holobionts we need to know how species assemble and interact. Controlled experiments are suited to study interactions between particular symbionts, but they only accommodate a tiny portion of the diversity within each species. Alternatively, interactions can be inferred by testing if associations among symbionts in the field are more or less frequent than expected under random assortment. However, random assortment may not be a valid null hypothesis for maternally transmitted symbionts since drift alone can result in associations. Here, we analyse a European field survey of endosymbionts in pea aphids (Acyrthosiphon pisum), confirming that symbiont associations are pervasive. To interpret them, we develop a model simulating the effect of drift on symbiont associations. We show that drift induces apparently nonrandom assortment, even though horizontal transmissions and maternal transmission failures tend to randomise symbiont associations. We also use this model in the approximate Bayesian computation framework to revisit the association between Spiroplasma and Wolbachia in Drosophila neotestacea. New field data reported here reveal that this association has disappeared in the investigated location, yet a significant interaction between Spiroplasma and Wolbachia can still be inferred. Our study confirms that negative and positive associations are pervasive and often induced by symbiont‐symbiont interactions. Nevertheless, some associations are also likely to be driven by drift. This possibility needs to be considered when performing such analyses, and our model is helpful for this purpose.