The aim of this paper is to compare different policy instruments for cost-effective habitat conservation on agricultural lands, when the desired spatial pattern of reserves is a random mosaic. We use a spatially explicit mathematical programming model which studies the farmers' behaviour as profit maximizers under technical and administrative constraints. Facing different policy measures, each farmer chooses its land-use at the field level, which determines the landscape at the regional level. A spatial pattern index (Ripley L function) is then associated to the obtained landscape, indicating on the degree of dispersion of the reserve. We compare a subsidy per hectare of reserve with an auction scheme and an agglomeration malus. We find that the auction is superior to the uniform subsidy both for cost-efficiency and for the spatial pattern of the reserve. The agglomeration malus does better than the auction for the spatial pattern but is more costly.