Biodiversity of soil is routinely assessed with environmental DNA—most often by massive parallel sequencing of marker genes (eDNA metabarcoding). Soil biodiversity may be investigated in relation to biodiversity research or as a tool in forensic investigations. After sampling, the taxonomic composition of soil biotic communities may change. In order to minimize community changes, it is desirable to reduce biological activity, e.g., by freezing immediately after sampling. However, this may be impossible due to remoteness of study sites or, in forensic cases, where soil has been attached to an item of interest for protracted periods of time. Here, we investigated the effect of storage duration and conditions on the assessment of the soil biota with eDNA metabarcoding. We extracted eDNA from freshly collected soil samples and again from the same samples after storage under contrasting temperature conditions and contrasting exposure (open/closed tubes). We used four different primer sets targeting bacteria, fungi, protists (cercozoans), and general eukaryotes. We quantified differences in richness, evenness, and community composition. Subsequently, we tested whether we could correctly infer habitat type and original sample identity after storage using a large reference dataset. We found stronger community composition differences with extended storage time and with higher storage temperature, and differences between open and closed tubes. However, for samples stored <28 days at a maximum of 20°C, changes were generally insignificant. Classification models successfully assigned most samples to their exact location of origin and correct habitat type even after 480 days storage. Even samples showing larger changes generally retained the original sample as the best match. For most biodiversity and forensic applications, storage of samples for days and even several weeks may thus not be a problem, if storage temperature does not exceed 20°C.