(1) Aerial survey methods for estimating population size of feral water buffalo in northern Australia were examined. (2) Line transect models underestimated population size. Of six models tested the most accurate underestimated by nearly half. The models give biased estimates most probably because not all animals on the survey line were sighted . (3) Aerial strip transect surveys were also negatively biased. The extent of this bias was estimated in index-removal experiments. Experiments were carried out on two populations in areas of differing obstructive canopy cover. (4) In woodland habitat with a canopy cover of 30- 60%, a correction factor of 3.2 was required to take account of animals not seen. In forest habitat with a canopy cover of 60-100%, a correction factor of 4.9 was required. (5) Using these results, the population size of feral buffalo, cattle and horses in Kakadu National Park was estimated by aerial survey at the end of each year over 6 years. Annual rates of increase for three regions of the Park were estimated, taking into account known removals from the population. The effects of dry season rainfall and population density in the preceding year on rate of increase were examined for each species. (6) The mean annual exponential rate of increase for each species was 0.10 yr-1 for buffalo, 0.23 yr-1 for cattle and -0.14 yr-1 for horses. (7) The annual rates of increase varied greatly between years within all species and were highly correlated with dry season rainfall in the year of survey for buffalo and cattle but not for horses. (8) No significant effect of preceding density on rate of increase was found for any species. A large reduction in buffalo populations did not correspond with an increase in unharvested populations of horses, suggesting the two species do not compete for food or other resources. (9) A campaign to control populations of feral water buffalo in Kakadu National Park was assessed. Between 1979 and 1988, approximately 79,000 animals were removed, 54% by commercial live-capture, 35% by shooting from helicopters and 10% by shooting from the ground. (10) In the period 1983-1988 when population estimates from aerial survey are available, mean buffalo population density was reduced from 5.60 km-2 to 1.17 km-2 over the surveyed area of the Park. (11) The costs of removal by shooting from helicopters, capturing animals alive and shooting from the ground were compared. The mean costs per animal in 1988 were $24.13, $74.53, and $86.02 respectively. (12) The effects of initial density and time spent shooting on number of animals removed by shooting from helicopters were examined. One linear and two curvilinear models were fitted to data from four different removal exercises. The relationship between time spent shooting and number removed was best described by a curvilinear (Ivlev) function. This model was used to estimate costs of control to a specified density. (13) Model regression coefficients differed between removal exercises, suggesting that the number removed may be affected by variables other than time spent shooting and initial density. Data from the range of conditions encountered during removal is thus likely to be required for robust estimation of removal costs.