Institutional changes in Benin have brought to light farmers' demand for varieties better suited to local growing conditions than existing ones. In response, we initiated a participatory cotton breeding experiment in 1996 to evaluate the relevance of such a methodology for the improvement of a commercial crop grown under rain-fed, semi-intensive cropping systems. This paper compares the performance of the first four mass-selection cycles, implemented by three farmer-breeders (F-B) and one formal breeder, with the original population and two commercial controls over three sites and two years. First results show that genetic changes occurred in all the F-B populations. The highest yielding F-B population (Savalou) was also more exuberant and later maturing than the others. Within the relatively narrow range of environments considered in the trial, there is no evidence that decentralized breeding results in better local adaptation. In Benin, participatory cotton breeding may be considered as complementary to formal on-station breeding and useful for enlarging the genetic variability offered to the farmers. Although the farmers want the approach to be scaled-up, its sustainability relies on a formal partnership between research and farmers institutions.