Ecological research in recent decades revealed that species loss has a predominantly negative effect on ecosystem functioning and stability. Most of these studies were based on random species loss scenarios, but extinctions in nature are not random. Recent experimental studies using macroscopic communities largely advanced knowledge about the effects of non-random species loss. However, in microscopic communities like the phytoplankton, implementing realistic species loss scenarios is challenging and experimental data are scarce. Creating more realistic experiments to study the role of phytoplankton diversity for ecosystem functioning is particularly important, as they provide up to 50% of global primary productivity, form the basis of all pelagic food webs, and are important for biogeochemical cycling. In this study, we experimentally tested and evaluated three methods for non-random species loss in a natural marine phytoplankton community. Dilution, filtration, and heat stress removed the targeted rare, large, and sensitive species, respectively. All these species groups are extremely vulnerable to extinction in future climate scenarios and play important roles in the communities. Dilution and filtration with a fine mesh additionally decreased initial biomass, which increased the variability of species left in the respective replicates. The methods tested in this study can be used to non-randomly manipulate phytoplankton species diversity in communities used for experiments. However, in studies where species identities are more important than species richness, the dilution and filtration methods should be modified to eliminate the effect of decreasing initial biomass.