Awareness of errors in health care has skyrocketed in recent years, and huge resources have been mobilised to measure and reduce the harm. This is a good thing, and long overdue. But current improvement recommendations have ignored the costs of prevention and have prioritized improvements by the rigour with which they have been studied. The current proliferation of safety goals and required or recommended safe practices threatens to overwhelm the capacity of hospitals to safely implement change, yet the cost-effectiveness of most proposed improvements remains unknown. Unless we collect information on cost-effectiveness, and use it to prioritize both improvement initiatives and new safety research, society will not gain the maximum return (in terms of safety) for whatever resources are put into error reduction. This would be a bad thing. Hospitals are complex systems, largely dependent on human performance, so improving hospital safety is not simple. Every change must be implemented with an understanding of human factors engineering and safety science, and even good changes can create unexpected new hazards. Increased safety precautions reduce preventable adverse events but generally impose both direct costs (to implement the safety precautions) and hidden costs (in the form of delays, new errors, or lost opportunities elsewhere). Perfect safety is not always possible and near-perfect-safety may impose unacceptably high costs. The goal of minimizing the total cost of both accidents and accident-prevention requires information on both costs and effects of specific safety improvements. Such information is also needed to prioritize suggested safety improvements, when all cannot be implemented immediately. This evidence can best be produced using the economic evaluation loop, an iterative process involving routine, periodic, assessment of costs and effects, and targeted original research where initial estimates reveal uncertainty in key values.