Headspace technique is a common method to analyze light hydrocarbons in deep marine sediments. The use of an automatic headspace sampler instead of a manual injection is the guarantee for a better repeatability and furthermore it is less time consuming. But to increase the sensitivity and therefore, to achieve a better detection limit, it is necessary to optimize the automatic headspace sampler parameters. The theory of design of experiments was applied here by studying them. As a response, methane, which is an important gas in marine sediments, was chosen for its short analysis time. Regarding the parameters for automatic headspace sampler, eight variables were selected and then, a screening of them was carried out with a fractional factorial design to determine the influential factors. Finally, optimization was conducted with four factors: Sample Loop Fill time (t(slf)), Oven temperature (T degrees(o)), Vial Pressurization time (t(vp)) and Vial pressure (P-v). They were modeled with a Doehlert experimental design. Then, the model was validated by a conventional statistical test (analysis of variance) and the optimum has been found and checked by three experiments. Results on light hydrocarbons measured in sediments from the Congo-Angola Basin are given, as an example. (C) 2008 Elsevier B.V. All rights reserved.