A major challenge in research into care at the end of life is the difficulty of obtaining the views and experiences of representative samples of patients. Studies relying on patients' accounts prior to death are potentially biased, as they only represent that proportion of patients with an identifiable terminal illness, who are relatively well and therefore able to participate, and who are willing to take part. An alternative approach that overcomes many of these problems is the retrospective or 'after death' approach. Here, observations are gathered from proxies, usually the patient's next of kin, following the patient's death. However, questions have been raised about the validity of proxies' responses. This paper provides a comprehensive review of studies that have compared patient and proxy views. The evidence suggests that proxies can reliably report on the quality of services, and on observable symptoms. Agreement is poorest for subjective aspects of the patient's experience, such as pain, anxiety and depression. The findings are discussed in relation to literature drawn from survey methodology, psychology, health and palliative care. In addition to this, factors likely to affect levels of agreement are identified. Amongst these are factors associated with the patient and proxy, the measures used to assess palliative care and the quality of the research evaluating the validity of proxies' reports. As proxies are a vital source of information, and for some patients the only source, the paper highlights the need for further research to improve the validity of proxies' reports.