As part of a programme to investigate the requirements for computer-based detection aids in the NHS breast screening programme, we have been studying how the work of mammogram reading is done within different screening centres. Our findings show how various interactional practices have been evolved in order to help maintain readers’ performance within acceptable levels. In this paper, we investigate one of these in detail, the use of annotations on screening reporting forms, and we consider their role in managing readers’ performance. Finally, we consider the implications of such practices for the design of computer-based systems and tools for breast screening.