The statistical analysis is reported of 256 paternity cases referred to seven different German laboratories for multilocus DNA fingerprinting with oligonucleotide probe (CAC)5/(GTG)5 and restriction enzyme HinfI. All parameters characteristic of multilocus DNA fingerprints were found to differ significantly between the contributing centres: the number of analyzed gel positions, the number of bands scored per individual, the probability of occurrence of a band at a particular position, and the band-sharing probabilities between the mother and both child and alleged father. Despite these differences, paternity cases could be divided clearly into two distinct subgroups on the basis of (i) offspring bands that could not be assigned to either the mother or the alleged father and (ii) the extent of band-sharing between child and alleged father. This partitioning, which is likely to correspond to true and false paternity, confirms previous findings for other multilocus probes. A goodness-of-fit test on the normalized number of bands scored per individual revealed no systematic deviations from commonly adopted analytical models regarding electrophoretic bands as independent entities. Log10-likelihood ratios of paternity vs. non-paternity were calculated utilizing one of these models, and a clear-cut partitioning was again obtained which coincides with that mentioned before. Only one case could not be decided unambiguously, and was either due to two independent mutations or to a close relative of the alleged father being the true father.