The increasing interest within development studies in religion is largely based on notions of 'faith communities' and 'belief systems': that peopleespecially 'religious' people -operate within discrete and coherent systems of belief. An emphasis on belief, however, is not universal, either across religions or across cultures. This paper draws on ethnographic data from a study of churches in rural Ghana to explore whether such frameworks are appropriate for understanding religious practices. Using insights from medical anthropology, it suggests that in this context the basis of 'religious' engagement is not belief as a conscious decision to adhere to a recognisably disputable notion. Rather, theoretical knowledge is preceded by practice, and continuity between the physical and the spiritual means that powers such as spirits are not 'believed in' (or disbelieved) but accepted as indisputable facts. Although people may identify with a particular religion such as Christianity, they live in a landscape of different 'religious' and 'non-religious' powers, with which they engage largely on a pragmatic level, entailing eclecticism, multiplicity and fluidity rather than full adherence to one discrete belief system. Thus, not only are assumed boundaries between religious groups and cosmologies challenged; but categories and oppositions used by development theoreticians and practitioners such as 'religious/secular' are also called into question.