<p>When local communities are assembled from a regional species pool, stochastic effects, such as differences in species arrival order and timing, can lead to alternative assembly outcomes. Here, we study the predictability of mutualistic network assembly using a trait-based model with explicit population dynamics. We vary size and composition of the species pool, immigration rates, and the strength of demographic noise. Assembly is more predictable when more species in the pool are only weakly dependent on their mutualistic partners or are interaction generalists. The size of the assembled networks increases with the pool size, with the proportion of species that are only weakly dependent on their mutualistic partner and with the immigration rate. The predominant assembly outcome at low immigration rates is a stable configuration where species composition does not change. At higher immigration rates, we observe also cycles (where a sequence of communities repeats periodically) and more complex temporal patterns. Our results are relevant for ecosystem restoration as they show how the composition of the regional species pool and the immigration rate influence the assembly outcome.</p>