This paper argues that if researchers studying a recognized category of behavior, such as fantasy play or aggression, were to focus on prototypical instances of the behavior, there should be high inter-researcher agreement of just the sort that operational definitions were intended to secure. A study of fantasy play was conducted using a paradigm case approach which focused on prototypical instances of fantasy play. The spontaneous interaction of pairs of young children was videotaped for a total of nearly 24 hours. Ninety paradigm cases (prototypical instances) of fantasy play were identified in the recorded interaction. The inter-observer reliability of the identification was tested. Phi coefficients were computed to compare each of four naive observers' judgments as to which portions of eight randomly selected film segments showed play pairs engaged in prototypical fantasy play. The average coefficient was .89, with a range of .84 to .94. It is concluded that the paradigm case approach secures high inter-researcher agreement without incurring a risk of experimental bias posed by the a priori construction of an operational definition of the phenomenon under study. (Author/RH)