This paper proposes a new approach for estimating the demand for seed within a developing country context where only improved seeds are sold but adoption rates for improved varieties low. A farmer views an improved seed firstly as a derived input embodying production attributes and secondly, as a technology embodying consumption characteristics. He therefore jointly decides on its adoption and the quantity of seed required to plant a predetermined area. Drawing on the theory of demand for consumption goods characteristics and production input attributes, this paper specified and estimated non-separable household demand and consumption models using data collected from 300 farm households in Zambia during the 2003/04 crop season. The estimated results suggest that adoption rate, distance to market, level of household grain self-sufficiency, seed hand-outs and household wealth are significant in determining farmers' seed purchase decisions. Appropriate intervention strategies for increased over-all improved seed demand are recommended. It is concluded that apart from contributing to the literature on modelling farm level seed demand, the model provides a holistic approach for the joint estimation of determinants of improved variety adoption and seed demand relevant for better targeting to increase the impacts of maize breeding research in developing countries.