A new approach is proposed to optimize some instrumental electrothermic parameters, i.e. injected volume, calcination duration, calcination and atomization temperatures, gas flow, in electrothermal atomic absorption spectrometry. It is based on the monitoring of the detection limit estimation variation by using an unconventional experimental design well adapted to a particular experimental domain. This method was tested on Cd, Mn and Cu in sea water. By comparison with the default operating conditions, the limits of detection were improved in some cases and always obtained on carrying out less experiments than usual. Copyright (C) 1998 Elsevier Science B.V.