Newborn deaths account for 57 of deaths in children younger than 5 years in Pakistan. Although a large programme of trained lady health workers (LHWs) exists, the effectiveness of this training on newborn outcomes has not been studied. We aimed to evaluate the effectiveness of a community-based intervention package, principally delivered through LHWs working with traditional birth attendants and community health committees, for reduction of perinatal and neonatal mortality in a rural district of Pakistan. We undertook a cluster randomised trial between February, 2006, and March, 2008, in Hala and Matiari subdistricts, Pakistan. Catchment areas of primary care facilities and all affiliated LHWs were used to define clusters, which were allocated to intervention and control groups by restricted, stratified randomisation. The intervention package delivered by LHWs through group sessions consisted of promotion of antenatal care and maternal health education, use of clean delivery kits, facility births, immediate newborn care, identification of danger signs, and promotion of careseeking; control clusters received routine care. Independent data collectors undertook quarterly household surveillance to capture data for births, deaths, and household practices related to maternal and newborn care. Data collectors were masked to cluster allocation; those analysing data were not. The primary outcome was perinatal and all-cause neonatal mortality. Analysis was by intention to treat. This trial is registered, ISRCTN16247511. 16 clusters were assigned to intervention (23 353 households, 12 391 total births) and control groups (23 768 households, 11 443 total births). LHWs in the intervention clusters were able to undertake 4428 (63) of 7084 planned group sessions, but were only able to visit 2943 neonates (24) of a total 12 028 livebirths in their catchment villages. Stillbirths were reduced in intervention clusters (39.1 stillbirths per 1000 total births) compared with control (48.7 per 1000; risk ratio [RR] 0.79, 95 CI 0.68-0.92; p=0.006). The neonatal mortality rate was 43.0 deaths per 1000 livebirths in intervention clusters compared with 49.1 per 1000 in control groups (RR 0.85, 0.76-0.96; p=0.02). Our results support the scale-up of preventive and promotive maternal and newborn interventions through community health workers and emphasise the need for attention to issues of programme management and coverage for such initiatives to achieve maximum potential. WHO; Saving Newborn Lives Program of Save the Children USA, funded by the Bill & Melinda Gates Foundation.