In most studies using informant reports of personality, the informants are recruited by the targets. Such informants may tend to like the targets and thus portray them in specific ways (e.g., too positively). Study 1 (N = 403) demonstrated the necessity to distinguish between "liking" and "knowing" in studying the relationships between informants and targets. Informants who liked their targets better described them more positively (i.e., as being more extraverted, agreeable, open, conscientious, and less neurotic). The association between personality ratings and liking was moderated by item evaluativeness. Liked targets were also described as being more similar to each other (range restriction), which would lower correlations of informant ratings of personality with external variables. Study 2 (N = 90) used a within-subjects design to disentangle actual personality differences between targets (substance) from the informants' specific views on the targets (style). Ninety informants were asked to nominate targets they did not like. These targets then nominated informants of their own choice. Target-nominated informants liked the targets very much and described the targets' personalities very positively and in ways that were highly redundant with the average target's self-rating. There was evidence that lower liking may be associated with better accuracy. Higher subjective levels of knowing were not associated with better interrater agreement but with a greater willingness to describe targets negatively. Implications for the field of recruiting are discussed. Where possible, the choice of informants and referees should be made independent of the targets' preferences.