The vastness of microbial diversity implies that an almost infinite number of individuals needs to be identified to accurately describe such communities. Practical and economical constraints may therefore prevent appropriate study designs. However, for many questions in ecology it is not essential to know the actual diversity but rather the trends among samples thereof. It is, hence, important to know to what depth microbial communities need to be sampled to accurately measure trends in diversity. We used three data sets of freshwater and sediment bacteria, where diversity was explored using 454 pyrosequencing. Each data set contained 6–15 communities from which 15 000–20 000 16S rRNA gene sequences each were obtained. These data sets were subsampled repeatedly to 10 different depths down to 200 sequences per community. Diversity estimates varied with sequencing depth, yet, trends in diversity among samples were less sensitive. We found that 1000 denoised sequences per sample explained to 90% the trends in β-diversity (Bray-Curtis index) among samples observed for 15 000–20 000 sequences. Similarly, 5000 denoised sequences were sufficient to describe trends in α-diversity (Shannon index) with the same accuracy. Further, 5000 denoised sequences captured to more than 80% the trends in Chao1 richness and Pielou's evenness.