Using environmental DNA (eDNA) to assess the distribution of micro- and macroorganisms is becoming increasingly popular. However, the comparability and reliability of these studies is not well understood as we lack evidence on how different DNA extraction methods affect the detection of different organisms, and how this varies among sample types. Our aim was to quantify biases associated with six DNA extraction methods and identify one which is optimal for eDNA research targeting multiple organisms and sample types. We assessed each methods' ability to simultaneously extract bacterial, fungal, plant, animal and fish DNA from soil, leaf litter, stream water, stream sediment, stream biofilm and kick-net samples, as well as from mock communities. Method choice affected alpha-diversity for several combinations of taxon and sample type, with the majority of the differences occurring in the bacterial communities. While a single method performed optimally for the extraction of DNA from bacterial, fungal and plant mock communities, different methods performed best for invertebrate and fish mock communities. The consistency of methods, as measured by the similarity of community compositions resulting from replicate extractions, varied and was lowest for the animal communities. Collectively, these data provide the first comprehensive assessment of the biases associated with DNA extraction for both different sample types and taxa types, allowing us to identify DNeasy PowerSoil as a universal DNA extraction method. The adoption of standardized approaches for eDNA extraction will ensure that results can be more reliably compared, and biases quantified, thereby advancing eDNA as an ecological research tool.