The paper examines both theoretically and empirically the factors determining the demand for regular labour in seasonal agriculture. In an implicit contract framework it is argued that there are 'hoarding costs' of regular labour in the slack period when there is not much work to be done. Consequently, the number of regular labour employed is constrained by the hoarding cost where larger employment-intensive farms tend to hire more regular labour. Evidence from the ICRISAT villages in India, too, show that though the marginal costs of regular labour are zero, there are significant hoarding costs of regular labour among small farms so that larger farmers are the major demanders of regular labour. Estimates of the double-hurdle model jointly determining the probability of hiring regular labour and demand for regular labour-hours (if a regular labour is hired) are shown to be an improvement over univariate tobit estimates of the demand for regular labour-hours only.