BACKGROUND: Evidence suggests that community-based interventions that promote improved home-based practices and care-seeking behaviour can have a large impact on maternal and child mortality in regions where rates are high. We aimed to assess whether an intervention package based on the WHO Integrated Management of Childhood Illness handbook and community mobilisation could reduce under-5 mortality in rural Guinea-Bissau, where the health service infrastructure is weak. METHODS: We did a non-masked cluster-randomised controlled trial (EPICS) in the districts of Tombali and Quinara in Guinea-Bissau. Clusters of rural villages were stratified by ethnicity and distance from a regional health centre, and randomly assigned (1:1) to intervention or control using a computerised random number generator. Women were eligible if they lived in one of the clusters at baseline survey prior to randomisation and if they were aged 15-49 years or were primary caregivers of children younger than 5 years. Their children were eligible if they were younger than 5 years or were liveborn after intervention services could be implemented on July 1, 2008. In villages receiving the intervention, community health clubs were established, community health workers were trained in case management, and traditional birth attendants were trained to care for pregnant women and newborn babies, and promote facility-based delivery. Registered nurses supervised community health workers and offered mobile clinic services. Health centres were not improved. The control group received usual services. The primary outcome was the proportion of children dying under age 5 years, and was analysed in all eligible children up to final visits to villages between Jan 1 and March 31, 2011. This trial is registered with ISRCTN, number ISRCTN52433336. FINDINGS: On Aug 30, 2007, we randomly assigned 146 clusters to intervention (73 clusters, 5669 women, and 4573 children) or control (73 clusters, 5840 women, and 4675 children). From randomisation until the end of the trial (last visit by June 30, 2011), the intervention clusters had 3093 livebirths and the control clusters had 3194. 6729 children in the intervention group and 6894 in the control group aged 0-5 years on July 1, 2008, or liveborn subsequently were analysed for mortality outcomes. 311 (4·6%) of 6729 children younger than 5 years died in the intervention group compared with 273 (4·0%) of 6894 in the control group (relative risk 1·16 [95% CI 0·99-1·37]). INTERPRETATION: Our package of community-based interventions did not reduce under-5 mortality in rural Guinea-Bissau. The short timeframe and other trial limitations might have affected our results. Community-based health promotion and basic first-line services in fragile contexts with weak secondary health service infrastructure might be insufficient to reduce child deaths. FUNDING: Effective Intervention.