In this thesis I explore the challenges for staff working to reduce harm and implement safety improvement in New Zealand (NZ) hospitals. Their views are contextualised in four stages. First, medical harm is outlined as a persistent and expensive threat to public health. While new practices make decisive action possible, implementation remains problematic. Second, policy in America, England, and NZ is analysed through Light’s (1995, 2010) theory of countervailing powers, and a shift from medical to managerial dominance. In NZ safety entered policy rhetoric around 2000, but it was compromised by resource shortages, a lack of evaluation, insufficient centralised support and coordination, and disengagement between managers and clinicians. While efforts intensified post-2008, resourcing remained problematic. Third, theories of organisational accidents (Reason 1990, 2000, 2001, 2004), normal accidents (Perrow 1984), sensemaking (Weick and Sutcliffe 2007), and the empirical literature about in-hospital risks, are reviewed. The review follows Vaughan’s (1999) discussion of organisational failure as emergent from the complex interconnection of organisational environments, organisations, and cognition and action. Pressure, organisational systems, hierarchy, communication, and organisational culture are identified as key risks. Fourth, the safety improvement literature is reviewed, implementation challenges are identified, and safety is theorised as emergent from unique organisational solutions to universal challenges of structure, culture, politics, learning, motivation, and infrastructure (Bate, Mendel et al. 2008). Staff perspectives from NZ are provided by n=37 qualitative interviews with doctors, nurses, and managers in three departments in two hospitals. Interviews explored the challenges of risk control and safety improvement, and are theorised as naturalistic accounts of real experiences. The dominant generalised risk was short staffing, which drove pressure, and contributed to poor communication and breakdowns in teamwork. These and other factors meant that some clinical risks were poorly controlled. Some denial of generalised and clinical risks was also evident. Improvement activities showed a number of gains, but many processes were failing from insufficient time, a lack of expertise in using systemic data and sensing problems, staff disengagement, poor ownership of processes, and inadequate IT infrastructure. In conclusion, tensions between productivity and safety pressured clinical work and contributed to ongoing harm. These failures drive up costs and threaten the fiscal sustainability of healthcare in NZ.