Recent work in child ethology has emphasized exhaustive descriptive studies of behavior. The concepts used as stimulus sets for counting have explicit reference to movements, postures, gestures, and expressions. These descriptive studies have generated a new set of questions concerning the construction and analysis of descriptive data sets. One of these questions concerns the assessment of interrater reliability. Some descriptors occur with such low frequency that an independent assessment of interrater reliability cannot be made. It is also clear that as the number of descriptors used in a particular study increases it becomes less feasible to independently assess interrater reliability for each descriptor. It is suggested that the average interrater reliability coefficient can be used to generate rules for systematically dealing with these problems.