In many cotton-producing countries, state companies organize the cotton sector, from seed production to lint marketing and extension. Most of them being privatised, their activities are taken over by a number of various actors. This may loosen the bonds between research and most stakeholders, including the farmers. To tighten these bonds, the Centre de coopération internationale en recherche agronomique pour le développement (Cirad) has been developing participatory approaches for ten years. Such approaches involve farmers in research and development activities. In particular, two projects of participatory selection have been developed. One was initiated in 1996 in Benin within the Institut national des recherches agricoles du Bénin (Inrab, National Institute of Agricultural Researches of Benin). The second one started in 2000 in Paraguay, with the Direction of Agricultural Research (DIA, Dirección de Investigación Agrícola) of the Ministry of Agriculture and Livestock. These programs were proposed to assess the relevance and the efficiency of two experimental and participatory designs. They consider two major aspects of cotton improvement: breeding and variety evaluation. In the decentralised scheme developed in Benin, an original and highly heterogenous popula-tion was developed by inter-crossing 14 varieties originated from various countries. From this population, seven samples were taken out and distributed to three farmer-breeders (F-B), one to the Inrab breeders and three were grown without conscious selection at. a close distance of the F-Bs' sites. For six years, the F-Bs selected these populations in the field, on their farms, and also from technological results obtained in the lab (ginning out turn and fibre characteristics). The Inrab breeders did the same on station. Having improved their skills, in 2003 the F-Bs could adopt pedigree instead of mass selection. In 2004, the best lines were compared on station, and also evaluated through the boll opening phase by farmers' representatives, F-Bs and Inrab breeders. Good lines were produced either by F-Bs or Inrab breeders. Some of them showed great improvement over the original population and they could match the commercial varieties. Most farmers elected four lines, two being produced by the F-Bs and two by the Inrab breeders. Those lines combined rather short internodes, numerous brandies loaded with well-opened bolls, and leaves falling at maturity. Farmers did not like later and more vegetative lines which were described as being "good for feeding cattle". When their crop was harvested and weighed, it happened that such late varieties could be more productive than some of the earlier ones which had first been preferred. Such promising results are not common, especially with a crop meant for industrial processing. They have convinced farmers' representatives, who are now concerned with the sustainability of the breeding program, and even with its scaling up. In Paraguay, public research is in charge of cotton genetic improvement. More participation is needed from the farmers to appraise their needs and preferences, to influence the breeding programs objectives, and to evaluate new varieties' performances before dissemination. A participatory variety evaluation design was set up in 2003-2004. Called "Base-Satellite" (Spanish)) tills design was directly inspired from the "Mother-Baby Trial". At each site, a "Base" trial compared the varieties in a replicated Fischer blocks design conducted by technicians. At the same site, five "Satellite trials" were planted with a subset of two new varieties and run by farmers-experimenters (F-Es) on their own farms. All through the growing season, F-Es could compare the new varieties with the commercial variety. They could express their opinions in questionnaires, and confront them during the field-days organised at each Base. Their opinions were compared with the variety performance in the multi-local and multi-annual national variety trial CNVT), The variety evaluation earned out by the F-Es' in the Satellites were highly correlated with its mean yield in the NVT (r2=0.708). The evaluation carried out in the Base alone was less strictly correlated (r2=.445). The design was particularly efficient in (a) pointing out the differences between on-station and on-farm evaluation, and (b) identifying possible defects before dissemination. It was particularly welcomed by the farmers, and could be the central piece of a participatory evaluation design in an organised sector. These two examples contribute to legitimate participatory approaches in cotton genetic improvement. They show how to re-create bonds between research and beneficiaries, when the coherence of the sector has been weakened after the national cotton companies were privatised.
