Contamination of drinking water supplies by arsenic is a world-wide problem. Total arsenic measurements are commonly used to investigate and regulate arsenic in water, but it is well understood that arsenic occurs in several chemical forms, and these exhibit different toxicities. It is problematic to use laboratory-based speciation techniques to assess exposure as it has been suggested that the distribution of species is not stable during transport in some types of samples. A method was developed in this study for the on-site speciation of the most toxic dissolved arsenic species: As (III), As (V), monomethylarsonic acid (MMA) and dimethylarsenic acid (DMA). Development criteria included ease of use under field conditions, applicable at levels of concern for drinking water, and analytical performance. The approach is based on selective retention of arsenic species on specific ion-exchange chromatography cartridges followed by selective elution and quantification using graphite furnace atomic absorption spectroscopy. Water samples can be delivered to a set of three cartridges using either syringes or peristaltic pumps. Species distribution is stable at this point, and the cartridges can be transported to the laboratory for elution and quantitative analysis. A set of ten replicate spiked samples of each compound, having concentrations between 1 and 60 μg/L, were analyzed. Arsenic recoveries ranged from 78% to 112% and relative standard deviations were generally below 10%. Resolution between species was shown to be outstanding, with the only limitation being that the capacity for As (V) was limited to approximately 50 μg/L. This could be easily remedied by changes in either cartridge design, or the extraction procedure. Recoveries were similar for two spiked hard groundwater samples indicating that dissolved minerals are not likely to be problematic. These results suggest that this methodology can be use for analysis of the four primary arsenic species of concern in drinking water supplies.