Human beings are influenced widely by relationship among individuals. However, there is still a lack of a systematic, objective and direct way to assess the quality of relationships within dyads. The current study proposed a task-specific method to evaluate relationship quality within friends and lovers. Thirteen-eight paired friends and 38 paired lovers were recruited. All of them performed the Pattern Matching and Go/No-go task, and their task performance were recorded. The friendship and love quality were measured respectively. Based on the results, there is significant influence relationship between the average step of initial level in Pattern Matching task and quality of friendship within friends. Furthermore, both average time of advanced level in Pattern Matching task and total RT difference in Go/No-go task were related to the relationship quality between lovers. The study demonstrates the validity of task-specific evaluation interface for different relationship types, and provides a method improving the understanding of human social relationships by machines, which may be effective for establishment of affective interaction between humans and computers.