This study examined children's aggressive behaviors in a classroom through observation, and also examined relationships between the observed aggressive behavior and self-reports, peer nomination, and teacher rating for aggressive behavior. Subjects were 36 5th grade elementary school children and their classroom behavior was recorded for 1970 minutes during 90 days. The frequency of observed aggressive behavior was occured total 120, that is, type I (reactive aggressor's physical aggression (PA) for initial aggressor's PA) 16, type II (reactive aggressor's verbal aggression (VA) for initial aggressor's PA) 49, type III (reactive aggressor's PA for initial aggressor's VA) 32, and type V (reactive aggressor's nonaggression for initial aggressor's PA and VA) 23. Aggressive behaviors from girl to boy were more than that from boy to girl. Correlations between observed frequencies of initial aggression and teacher rating for an initial aggression, between observed frequencies of reactive aggression and teacher rating for an initial aggression, self-report for an initial aggression were significant. From these correlations prediction of aggressive behaviors were considered.