The unique holotype of Hallopus victor (Marsh), from the Upper Jurassic of Garden Park, Colorado, is redescribed. The bones previously identified as pubes (Marsh 1890) or ischia (von Huene 1914) are regarded here as the left radius and ulna, and the 'ulna' and 'radius' of previous workers are considered to be the left radiale and ulnare. Marsh's identification (1890) of the ischium and his orientation of the scapula and femur (1896) are upheld. The presence of a humerus on the larger slab is confirmed. Hallopus is interpreted as a highly specialized, cursorial crocodilian, with slender, hollow bones, a greatly elongated radiale and ulnare, and a roller-like joint between these and the metacarpals. The manus is pentadactyl with a symmetrical distribution of lengths about the central axis and some proximal wedging-out of the metacarpals. The iliac blade is elongated and resembles that of Orthosuchus, the ischium is reminiscent of that of Protosuchus. The femur has a lesser trochanter, a fourth trochanter and a 'pseudointernal' trochanter, but no greater trochanter. The tibia is longer than the femur. The tarsus is basically crocodilian in pattern, but greatly compressed and specialized. The first metatarsal is reduced to an elongated splint, permanently recessed into metatarsal II. Metatarsals II to IV are symmetrical in length with III longest, metatarsal V is reduced, pointed, and lacks phalanges. The interpretation put forward provides a consistent explanation of the peculiarities of the skeleton of Hallopus as a variant on the basic crocodilian plan. The details of the articulation of the carpal and tarsal joints are described as far as preservation permits, and possible movements are considered. The carpometacarpal and tarsal joints are simple hinges, but the proximal carpal joint appears to have been relatively immobile and the elongation of the radiale and ulnare is viewed as a device to compensate for the increase in length of the tibia. The femur has an off-set, ball-like head and evidently moved essentially in a parasagittal plane. The pes is functionally tridactyl, with the metatarsals locked together proximally. It is concluded that both fore- and hind-feet were digitigrade during movement, although in a stationary pose the metatarsus may have been in contact with the ground. Some aspects of the pelvic and hind-limb musculature are briefly discussed. Functional analogies from the locomotory point of view are limited by the lack of cursorial quadrupedal archosaurs for comparison. It is concluded that a hare-like bounding gallop was the most probable type of fast locomotion in Hallopus. Although no skull bones have been identified, evidence from the postcranial skeleton is adduced to show that Hallopus is of pedeticosaurid descent. The relationships of early crocodilomorphs are discussed, and it is deduced that two basic stocks diverged from a common ancestry during the middle part of the Trias. These two groups are included in an expanded Order Crocodylomorpha. The Suborder Crocodylia has the Triassic Stegomosuchidae as its radicle and contains 'normal' crocodiles (including the Sebecosuchia but not the Baurusuchidae). The suborder Paracrocodylia is proposed for mainly cursorial forms, to include the infraorders Pedeticosauria, Baurusuchia and Hallopoda. Diagnoses for these groupings are presented. An origin for both stocks from a form close to Cerritosaurus is postulated. Erpetosuchus and Dyoplax are not now regarded as crocodilomorphs. The possibility of an early cursorial phase in crocodilian evolution is briefly discussed, and it is tentatively suggested that the gallop occasionally observed in young crocodiles (Cott 1961) may be a relic of a primitive type of locomotion in the group. The significance of this to the emergence of the crocodilian type of shoulder-girdle is considered.