This paper argues that, for the rural poor in Africa, market failure is more the norm than the exception. Despite the growing attention given to market imperfections of the kind highlighted by New Institutional Economics, much policy advice on the agricultural economy in African countries remains based on unrealistic analysis and assumptions. To make policy advice more relevant requires a better understanding not only of how markets (mal)function, but also of implementation issues: what constrains implementability, how constraints can be overcome or bypassed, and what policy measures have greater or lesser prospects of implementation.It concludes that: The policy discourse in African agriculture has been dominated by agricultural economics, yet many of the standard assumptions of quantitative modelling are challenged by the reality of:; markets which are routinely non-competitive because of, eg. high barriers to entry, increasing returns to scale, and non-insurable risk; deliberate efforts by firms to maintain market imperfections by, e.g. building information exchange networks which exclude newcomers; the absence of market pressures towards more competitive structure, conduct or performance The major task of including the poor more fully in markets will continue to be neglected for as long as policy advice remains dominated by unrealistic assumptions; concerning the structure, conduct and performance of markets; This discourse has tended to ignore political priorities such as achieving (or denying) developmental balance among regions or tribes; In addition, the discourse has generally sought some ideal of policy design, disregarding questions of implementability. Much more attention is needed to implementation constraints, such as weak infrastructure, power relations, and corruption, and to how these can be overcome or circumvented.; If evidence-based research is to be more relevant to policy, it will have to be re-tooled to make good the above deficiencies. In particular, there is opportunity for new methods which are more qualitative, such as case study or networking analysis. Inevitably, these will also require a broader mix of disciplines than agricultural economics, which has been dominant hitherto.