Partial least-squares (PLS) regression calibration was applied to the determination of arsenic by stripping potentiometry on gold electrodes. In a procedure involving the direct electrochemical reduction of arsenic (V), this allowed interferences from copper (II), due to formation of intermetallic compounds, and tin (IV), due to stripping peak overlaps, to be modeled. At levels of arsenic below 1 mug L-1, additional problems with the gold surface background, as well as copper interferences from a reagent blank, could be successfully modeled, and the prediction of sample concentrations could be accurately performed.