Although studies examining changes in the structure and diversity of microbial communities in biological soil crusts (BSCs) have increased, microbial interactions in BSCs are not currently fully understood. In this study, we applied a random matrix theory (RMT)-based network approach to construct bacterial, fungal and bacteria-fungal interaction networks for the four developmental stages of BSCs (bare sand, physical crusts, algae crusts and moss crusts) in the Mu Us Sandy Land, northwestern China. Our results showed that Firmicutes, Proteobacteria, Cyanobacteria and Actinobacteria were the keystone phylum in bacterial networks in the four developmental stages, respectively. Ascomycota was the keystone phylum in the fungal networks in all stages, and Basidiomycota was a further keystone phylum in the algae and moss crust networks. With the development of BSCs, complexity in bacterial and fungal networks increased, becoming more clustered and connected, indicating a strengthened pattern in microbial interactions with BSC development. Positive links dominated in all networks, showing that microbial synergism was important in the development of BSCs. Interestingly, with the development of BSCs, the proportion of negative links in bacterial networks increased and they decreased in fungal networks, suggesting more inter-community competition was present in the bacterial communities and less inter-community competition was present in the fungal communities. Environmental factors, especially total carbon and total organic carbon, constrained most bacterial and fungal communities, respectively. Our findings provided a new insight into understanding BSC developmental mechanisms and ecological functions in a desert ecosystem composed of a sandy land landscape in arid and semi-arid regions.