By initiating a participatory cotton breeding experiment in 1996, researchers have designed new technologies to adapt to the diversity of local farming conditions. The study compared the performance of farmer selected populations derived after 4 cycles of recurrent mass selection. These populations were evaluated along with 2 commercial checks, STAM 18 A and H-279-1 and the original population AGP 0 at five locations, Angaradebou, Mone, Savalou, Okpara in Benin republic and Samaru in Nigeria in two years. In comparison with the average of the local checks, the farmers made significant improvement in population density, days to boll opening, seed cotton yield and seed index. The average of the populations from farmer's selection was compared to the average of the research selection. The results showed that there were significant differences between farmer's selection and research selection with respect to seed cotton yield (68.5 kg/ha) and lint yield (34.2 kg/ha). These characters were improved by the farmers. During years of experiment, the farmers worked with enthusiasm and were able to select promising cotton populations with good parameters. The final genetic material (lines) will soon be available for on farm testing in more contrasting cropping systems and environments.