The purpose of the present study was to clarify the method of estimating responsive behaviors of profoundly retarded persons. Subjects were 3 profoundly retarded persons (DA 0: 4-1: 5). The stimuli were human faces and red lights. Their responsive behaviors were recorded by VTRs. The three observers judged and marked occurrence of behaviors. About the reliability of observation, a coefficient of interobserver agreement (Kappa) was calculated. Responsive behaviors were examined through calculating the conditional probability of occurrence of behaviors on the condition of presenting a stimulus. Results indicated that a good accordance was found between observers about the time-course of conditional probability of the behavior, when a coefficient of interobserver agreement was more than 0.7 that was siginificantly different from chance agreement (FIG. 2). About characteristics of responses, the time-course of conditional probability indicated the higher responsiveness to stimuli of human faces than red lights in each retarded persons (FIG. 3). From the above mentioned results, we pointed out that the present method might be effective in examining responsive behaviors of profoundly retarded persons.