Self-reports and spouse ratings of personality traits typically show less-than-perfect agreement, but powerful moderators of agreement have not yet been identified. In Study 1, 47 married couples completed the Revised NEO Personality Inventory to describe themselves and their spouses. Extent of agreement was not consistently moderated by response sets; the age, intelligence, or education of the respondent; or the length or quality of the relationship. In Study 2 these couples were interviewed about reasons for substantial disagreements, and an audiotape was content-analyzed. Sixteen reasons were reliably coded, including idiosyncratic understanding of items, reference to different time frames or roles, and unavailability of covert experience to the spouse. Faking good, assumed similarity, and other variables prominent in the psychometric literature were relatively unimportant. Findings (1) suggest that attempts to improve the validity of self-reports and ratings may need to be refocused and (2) underscore the desirability of routinely obtaining multiple sources of information on personality.