In this paper, a multivariable approach has been applied for the selective removing of Fe(III) and Al(III), in the range 0-200 mu g l(-1), in water samples onto a modified organic support (salicylic acid grafted on XAD-4). An empirical mathematical model was designed which establishes the relationship between the variation of the responses (extraction yields), and the variation of three factors (sample volume, sample percolation flow rate and amount of metallic ions present in the sample). To estimate the coefficients of the developed model, an uniform shell Doehlert design has been applied; these experiments consisted in GF-AAS determination of aluminium and iron amounts in eluates after percolation of samples through modified support. Results show a similar behaviour of the resin towards aluminium and iron with a preponderant effect of the percolation flow rate value; however this one is crucial for aluminium extraction and should be maintained below to 0.55 ml min(-1) to reach a 95% Al3+ extraction yield (versus 2.25 ml min(-1) for Fe3+). The optima determined by this experimental design approach have been further applied to the selective extraction of aluminium and iron from multielement synthetic samples and from real samples at the outlet of potable water treatment units. (C) 2007 Elsevier B.V. All rights reserved.