- The Siwalik formations of northern Pakistan consist of deposits of ancient rivers that existed throughout the early Miocene through the late Pliocene. The formations are highly fossiliferous with a diverse array of terrestrial and freshwater vertebrates, which in combination with exceptional lateral exposure and good chronostratigraphic control allows a more detailed and temporally resolved study of the sediments and faunas than is typical in terrestrial deposits. Consequently the Siwaliks provide an opportunity to document temporal differences in species richness, turnover, and ecological structure in a terrestrial setting, and to investigate how such differences are related to changes in the fluvial system, vegetation, and climate. Here we focus on the interval between 10.7 and 5.7 Ma, a time of significant local tectonic and global climatic change. It is also the interval with the best temporal calibration of Siwalik faunas and most comprehensive data on species occurrences. A methodological focus of this paper is on controlling sampling biases that confound biological and ecological signals. Such biases include uneven sampling through time, differential preservation of larger animals and more durable skeletal elements, errors in age-dating imposed by uncertainties in correlation and paleomagnetic timescale calibrations, and uneven taxonomic treatment across groups. We attempt to control for them primarily by using a relative-abundance model to estimate limits for the first and last appearances from the occurrence data. This model also incorporates uncertainties in age estimates. Because of sampling limitations inherent in the terrestrial fossil record, our 100-Kyr temporal resolution may approach the finest possible level of resolution for studies of vertebrate faunal changes over periods of millions of years. Approximately 40,000 specimens from surface and screenwash collections made at 555 localities form the basis of our study. Sixty percent of the localities have maximum and minimum age estimates differing by 100 Kyr or less, 82% by 200 Kyr or less. The fossils represent 115 mammalian species or lineages of ten orders: Insectivora, Scandentia, Primates, Tubulidentata, Proboscidea, Pholidota, Lagomorpha, Perissodactyla, Artiodactyla, and Rodentia. Important taxa omitted from this study include Carnivora, Elephantoidea, and Rhinocerotidae. Because different collecting methods were used for large and small species, they are treated separately in analyses. Small species include insectivores, tree shrews, rodents, lagomorphs, and small primates. They generally weigh less than 5 kg. The sediments of the study interval were deposited by coexisting fluvial systems, with the larger emergent Nagri system being displaced between 10.1 and 9.0 Ma by an interfan Dhok Pathan system. In comparison to Nagri floodplains, Dhok Pathan floodplains were less well drained, with smaller rivers having more seasonally variable flow and more frequent avulsions. Paleosol sequences indicate reorganization of topography and drainage accompanying a transition to a more seasonal climate. A few paleosols may have formed under waterlogged, grassy woodlands, but most formed under drier conditions and more closed vegetation. The oxygen isotopic record also indicates significant change in the patterns of precipitation beginning at 9.2 Ma, in what may have been a shift to a drier and more seasonal climate. The carbon isotope record demonstrates that after 8.1 Ma significant amounts of C4 grasses began to appear and that by 6.8 Ma floodplain habitats included extensive C4 grasslands. Plant communities with predominantly C3 plants were greatly diminished after 7.0 Ma, and those with predominantly C4 plants, which would have been open woodlands or grassy woodlands, appeared as early as 7.4 Ma. Inferred first and last appearances show a constant, low level of faunal turnover throughout the interval 10.7–5.7-Ma, with three short periods of elevated turnover at 10.3, 7.8, and 7.3–7.0 Ma. The three pulses account for nearly 44% of all turnover. Throughout the late Miocene, species richness declined steadily, and diversity and richness indices together with data on body size imply that community ecological structure changed abruptly just after 10 Ma, and then again at 7.8 Ma. Between 10 and 7.8 Ma the large-mammal assemblages were strongly dominated by equids, with more balanced faunas before and after. The pattern of appearance and disappearance is selective with respect to inferred habits of the animals. Species appearing after 9.0 Ma are grazers or typical of more open habitats, whereas many species that disappear can be linked to more closed vegetation. We presume exceptions to this pattern were animals of the mixed C3/C4 communities or the wetter parts of the floodplain that did not persist into the latest Miocene. The pace of extinction accelerates once there is C4 vegetation on the floodplain. The 10.3 Ma event primarily comprises disappearance of taxa that were both common and of long duration. The event does not correlate to any obvious local environmental or climatic event, and the pattern of species disappearance and appearance suggests that biotic interactions may have been more important than environmental change. The 7.8 Ma event is characterized solely by appearances, and that at 7.3 Ma by a combination of appearances and disappearances. These two latest Miocene events include more taxa that were shorter ranging and less common, a difference of mode that developed between approximately 9.0 and 8.5 Ma when many short-ranging and rare species began to make appearances. Both events also show a close temporal correlation to changes in floodplain deposition and vegetation. The 7.8 Ma event follows the widespread appearance of C4 vegetation and is coincident with the shift from equid-dominated to more evenly balanced large-mammal assemblages. The 7.3 to 7.0 Ma event starts with the first occurrence of C4-dominated floras and ends with the last occurrence of C3-dominated vegetation. Absence of a consistent relationship between depositional facies and the composition of faunal assemblages leads us to reject fluvial system dynamics as a major cause of faunal change. The close correlation of latest Miocene species turnover and ecological change to expansion of C4 plants on the floodplain, in association with oxygen isotopic and sedimentological evidence for increasingly drier and more seasonal climates, causes us to favor explanations based on climatic change for both latest Miocene pulses. The Siwalik record supports neither “coordinated stasis” nor “turnover pulse” evolutionary models. The brief, irregularly spaced pulses of high turnover are characteristic of both the stasis and pulse models, but the high level of background turnover that eliminates 65–70% of the initial species shows there is no stasis in the Siwalik record. In addition, the steadily declining species richness and abrupt, uncoordinated changes in diversity do not fit either model.