Food assistance in the form of cash has increased in recent years, in part due to proven cost and timeliness advantages relative to food transfers. Moreover, economic theory generally predicts that providing greater choice should improve welfare, such that food consumption outcomes should be the same or better with cash than with food transfers. Empirical investigations have shown, however, that people sometimes prefer to receive food, and that the marginal propensity to consume food is often greater from in-kind than from cash transfers. The mechanisms and rationale behind the conditional superiority of food transfers is not fully understood. This paper presents a model of household decisionmaking that integrates cash versus food transfers, and frictions (transaction costs), to generate predictions about the use of transfers, acknowledging the infra- or extra- marginality of different components of the food basket. The model predicts that when food is extra-marginal a food transfer of an inexpensive staple grain can have negative impacts on dietary diversity, relative to a cash transfer, but positive impacts for higher quality food transfers. These differences are primarily due to the transaction costs involved in selling food that lower the shadow prices of extra-marginal transferred foods. I find support for this model using data from a randomized cash-food pilot project in Eastern Niger.