This study analyses the referral patterns of patients, over time, from primary care to secondary or tertiary level facilities in rural Africa. The data come from a health information system of a non-governmental organization with a decade of experience in health services delivery in Samburu District, Kenya. The differential referral patterns from two communities are examined in some detail to shed more light on the meaning of a 'referral rate' in this context. First, referral rates over time for two clinics are calculated and compared. These quantitative data, obtained from monthly reports from 1989 to 1997, are interpreted in the light of qualitative data obtained from interviews with community health workers, nurses and members of the communities. The main differences in referral between these ostensibly similar communities are for malaria, trauma and anaemia. Social, environmental and specific health services factors are used to explain these differences. We demonstrate that basic information from health information systems, which tell us little on their own because referral in this context is a rare event, can be combined with local knowledge from the community to provide evidence for health managers to set priorities for public health and clinical interventions.