The incorporation of precise definitions for taxonomic units into wildlife legislation has necessitated the reevaluation of the taxonomy of endangered and threatened species. We used the subspecies recognition criteria proposed by Avise and Ball (1990) and O’Brien and Mayr (1991) to examine the infraspecific taxonomy of the leopard, Panthera pardus, a geographically widespread species with 27 currently recognized trinomial designations. Samples from named subspecies revealed appreciable genetic diversity using three molecular methods: allozymes, mitochondrial DNA restriction sites, and feline‐specific minisatellites. Continental populations and subspecies from Africa and Asia possessed the highest amount of molecular genetic variation, whereas relatively lower amounts of diversity were present in island populations. Molecular data were analyzed using three phylogenetic methods (distance‐matrix, maximum parsimony, and maximum likelihood) to resolve genetic differentiation below the species level The combined results revealed phylogenetic distinction of six geographically isolated groups of leopards: (1) African, (2) central Asian, (3) Indian, (4) Sri Lankan, (5) Javan, and (6) east Asian. Based on the combined molecular analyses and supporting morphological data (Miththapala 1992), u,e recommend that subspecific leopard taxonomy be revised to comprise eight subspecies: (1) P. p. pardus, Africa; (2) P. p. saxicolor, central Asia; (3) P. p. fusca, Indian subcontinent; (4) P. p. kotiya, Sri Lanka; (5) P. p. melas, Java; (6) P. p. orientalis, Amur; (7) P. p. japonensis, northern China; and (8) P. p. delacouri, southern China. In most cases, designated subspecies conform to historic geological barriers that would have facilitated allopatric genetic divergence.

(With 2 figures in the text)Twenty-one captive leopards (Panthera pardus kotiya) at the National Zoological Gardens in Sri Lanka were individually identified using spot pattern variation. Based on an identification method established for lions {Panthera leo), a code was devised examining 23 variable characters, each of which had one to three values. These characters ranged from number and spacing of muzzle spots to forehead and eye patterns. Correlation among characters to be used for identification was minimized using principal component cluster analysis. The most variable character was chosen from each of eight non-overlapping clusters, and frequencies were calctilated for each character value. The probabiUty of occurrence of a given spot pattern was calculated as the sum of the frequencies of each character value. From this probability of occurrence, the information content, in bits, was computed for each pattern. The number of bits per character was also calculated. Using the binomial theorem, the rehability of identification was estimated as the sum of the probabilities of zero or one individual having an identical combination of character values. These binomial probabilities exceeded 0-99 for 15 out of 21 animals, and 0-95 for all but two. In these two animals, the information content was low (5-99 and 5-50 bits, respectively) compared to the others, in which information content ranged from 6-87-10-86. Although the mean number of bits (8-5) was theoretically sufficient for a 99% reliable identification, it was concluded that supplementation with an additional character would be worthwhile for identification of individual leopards.