In recent years, methodological advances have substantially improved our ability to recover DNA molecules from ancient samples, raising the possibility to sequence palaeogenomes without PCR amplification. Here we present an amplification-free library preparation method based on a benchmark library preparation protocol in palaeogenomics based on single-stranded DNA, and demonstrate suitability of the new method for a range of sample types. Furthermore, we use the method to generate the first amplification-free nuclear genome of a Pleistocene cave bear, and analyse the resulting data in the context of cave bear population genetics and phylogenetics using standard genomic clustering analyses. We find that the PCR-free adaptation provides endogenous DNA contents, GC contents and fragment lengths consistent with the standard protocol, although with reduced conversion efficiency, and shows no biases in downstream population clustering analyses. Our amplification-free library preparation method could find application in experimental designs where the original template molecule needs to be characterised more directly.