This paper argues that contextual constraints can significantly interfere with the conduct of research and development, and accordingly should be analyzed more honestly. It describes a five country research project, Landscape Mosaics, in which participatory action research (PAR) was intended as one of two central approaches in the original research design (the other approach being more conventional research). The five sites, in Cameroon, Indonesia, Laos, Madagascar, and Tanzania, are described, with an emphasis on their implementation of PAR. The fact that personnel and partners on three of the five sites failed to implement PAR is analyzed, to determine the constraints to such action. These findings are then compared with our experience with two other similar projects in which PAR was more widely implemented. We conclude by identifying the most important constraints to be overcome in implementing a PAR process, something we consider important in efforts to deal with change processes in developing countries.