Numerous studies comparing patients' end-of-life health care treatment preferences with their surrogates' reports of those preferences indicate that partners do not know one another's treatment preferences: Random guesses are just as likely as surrogate choices to match the patients' preferences. The present study uses the empathic accuracy model and the assumed similarity model to shed further light on accuracy and bias in surrogate reports. The data come from the National Center for Family and Marriage Research/Knowledge Networks Pilot Study 2010, a national sample of 1,075 heterosexual American couples age 18 to 64. Surrogate reports were simultaneously biased (i.e., correlated with the surrogate's own preference) and accurate (i.e., correlated with the partner's actual preference). Higher relationship quality and legal marriage (vs. cohabitation) were associated with increased assumed similarity bias. These findings suggest that practitioners working in end-of-life care need to be familiar with their partnered patients' relationship dynamics.