Fewstudies have assessed the effective gains achieved by collaborative decentralized participatory breeding, in terms of yield and other agronomic traits, in comparison with centralized conventional breeding using the same breeding schemes. Our study concerned an experience of participatory improvement of tortillero sorghum for low-input cropping systems in northern Nicaragua. It set out to compare the effect of two breeding schemes simultaneously managed on-station and on-farm. Two synthetic populations were used as sources of genetic variability. The study was designed to distinguish between three "selection modes": farmers' selection on-farm (FoF), breeder' selection on-station (BoS) and breeder's selection on-farm (BoF). After two selection cycles, we found that FoF selection reduced phenotypic variability as much as BoS selection. In general, BoS selection produced higher-yielding lines under the target on-farm conditions. But FoF selection was more effective in promoting lines with higher values for an index of agronomic suitability (IAS), expressing a better balance between earliness, plant height, grain size and yield. Such genotypes might cope more easily with local environmental constraints and farmers' preferences. BoF selection did not prove to be effective for either yield or IAS. This research shows the need for close collaboration between farmers and breeders as well as complementarities between the selection phases managed on-farm and on-station, to make breeding for difficult environments fully efficient.