A volume-based flow injection (FI) procedure is described for the determination and speciation of trace inorganic arsenic, As(III) and As(V), via hydride generation-atomic absorption spectrometry (HG-AAS) of As(III). The determination of total arsenic is obtained by on-line reduction of As(V) to As(III) by means of 0.50% (w/v) ascorbic acid and 1.0% (w/v) potassium iodide in 4 M HCl. The combined sample and reduction solution is initially heated by flowing through a knotted reactor immersed in a heated, thermostatted oil bath at 140 degrees C, and subsequently, for cooling the reaction medium, a knotted reactor immersed in a water bath at 10 degrees C. By using the very same volume-based FI-HG-AAS system without the heating and cooling reactors, and employing mild hydrochloric acid conditions, As(V) is not converted to arsine, thereby allowing the selective determination of As(III). The injected sample volume is 100 mu l while the total sample consumption per assay is 1.33 ml, and the sampling frequency is 180 samples per hour. The detection limit (3 sigma) for the on-line reduction procedure was 37 ng l(-1) and at the 5.0 mu g l(-1), the relative standard deviation (RSD) was 1.1% (n=10) by calibrating with As(III) standards; by calibrating with As(V) standards the detection limit was 33 ng l(-1) and the RSD was 1.3% (n=10). For the selective determination of As(III) the detection limit was 111 ng l(-1) and the RSD was 0.7% (n=10) at 5.0 mu g l(-1). Both procedures are most tolerant to potential interferents. Thus, without impairing the assay, interferents such as Cu, Co, Ni and Se could, at a As(V) level of 5 mu g l(-1), be tolerated at a weight excess of 2000, 30000, 200 and 200 times, respectively. The assay of a certified drinking water sample by means of multiple standard addition (five levels; each three replicates) was 9.09+/-0.05 mu g l(-1) (certified value 9.38+/-0.71 mu g l(-1)).