BACKGROUND: Effective prevention and control of diarrhoea requires caregivers to comply with a suite of proven measures, including exclusive breastfeeding, handwashing with soap, correct use of oral rehydration salts, and zinc administration. We aimed to assess the effect of a novel behaviour change intervention using emotional drivers on caregiver practice of these behaviours. METHODS: We did a cluster randomised controlled trial in Lusaka Province, Zambia. A random sample of 16 health centres (clusters) were selected from a sampling frame of 81 health centres in three of four districts in Lusaka Province using a computerised random number generator. Each cluster was randomly assigned 1:1 to either the intervention-clinic events, community events, and radio messaging-or to a standard care control arm, both for 6 months. Primary outcomes were exclusive breastfeeding (self-report), handwashing with soap (observation), oral rehydration salt solution preparation (demonstration), and zinc use in diarrhoea treatment (self-report). We measured outcome behaviours at baseline before start of intervention and 4-6 weeks post-intervention through repeat cross-sectional surveys with mothers of an infant younger than 6 months and primary caregivers of a child younger than 5 years with recent diarrhoea. We compared outcomes on an intention-to-treat population between intervention and control groups adjusted for baseline behaviour. The study was registered with ClinicalTrials.gov, number NCT02081521. FINDINGS: Between Jan 20 and Feb 3, 2014, we recruited 306 mothers of an infant aged 0-5 months (156 intervention, 150 standard care) and 343 primary caregiver of a child aged 0-59 months with recent diarrhoea (176 intervention, 167 standard care) at baseline. Between Oct 20 to Nov 7, 2014, we recruited 401 mothers of an infant 0-5 months (234 intervention, 167 standard care) and 410 primary caregivers of a child 0-59 months with recent diarrhoea (257 intervention, 163 standard care) at endline. Intervention was associated with increased prevalence of self-reported exclusive breastfeeding of infants aged 0-5 months (adjusted difference 10·5%, 95% CI 0·9-19·9). Other primary outcomes were not affected by intervention. Cluster intervention exposure ranged from 11-81%, measured by participant self-report with verification questions. Comparison of control and intervention clusters with coverage greater than 35% provided strong evidence of an intervention effect on oral rehydration salt solution preparation and breastfeeding outcomes. INTERPRETATION: The intervention may have improved exclusive breastfeeding (assessed by self-reporting), but intervention effects were diluted in clusters with low exposure. Complex caregiver practices can improve through interventions built around human motives, but these must be implemented more intensely. FUNDING: Absolute Return for Kids (ARK) and Comic Relief.