Biological dinitrogen fixation (BNF) represents a major input of reduced nitrogen (N) to the oceans. Accurate direct measurements of BNF rates are crucial for reliably determining the biogeochemical significance of diazotrophy at local and global scales. Traditionally, borosilicate glass fiber filters (GF/F, Whatman) with a nominal pore size of 0.7 μm are used to collect suspended particles by filtration after incubations with added 15N2 tracer. We carried out BNF experiments in the Baltic Sea, Danish coastal waters, and the Pacific Ocean comparing the retentive characteristics of precombusted GF/F filters with newer Advantec glass fiber filters which have a smaller nominal pore size of 0.3 μm. Where BNF was detected, rates were nearly always higher, and sometimes even exclusively detectable, when using Advantec filters. In the majority of samples across tested habitats, significantly more cells were lost to GF/F filtrate (average = 51%, range = 10–70% of cells) than to Advantec filtrate (average = 40%, range = 10–54%). Using Illumina sequencing of nitrogenase (nifH) gene amplicons, we show that diazotroph communities can markedly differ between bulk water and filtrates from GF/F and Advantec filtrations, suggesting that different diazotrophs can pass through the filter types. In order to reduce the potential underestimations of BNF due to filtration loss of diazotrophs, we recommend using Advantec filters or alternatively silver membranes with 0.2 μm pore size, especially in waters expected to be inhabited by relatively small, unicellular diazotrophs.