It follows from previous studies that changes in the base pair vertical separation (BPVS) influence the architecture of DNA much more than any other conformational parameter. This inspired us to compare BPVS in the available oligonucleotide crystal structures with the optimum values provided by nine different empirical potentials employed in the theoretical studies of DNA conformation. This comparison shows that BPVS is reproduced by three fields in all steps of the highly resolved oligonucleotide crystal structures while the remaining six empirical potentials, including AMBER, GROMOS and CHARMM, provide systematic deviations. We further find that the base pairs are poorly stacked (mostly compressed) in some other refined DNA crystal structures. Our analysis indicates that this poor stacking originates from improperly determined positions of the bases. The approach described in the present communication can be used to identify DNA structures which are not accurate enough for studies of the relationships between the base sequence and DNA conformation.