This study compared three ways of analyzing individual mother-infant attachment behaviors in order to test the hypothesis that success in the search for stable individual differences in attachment behavior is in part a function of the level at which behavior individuality is assessed. Fifty infants were videotaped in the Ainsworth and Wittig Strange Situation at 12 and 18 months of age. Three different assessments of individual differences were employed: (1) time samples of discrete behaviors directed toward adults (look, glance, vocalize, etc.); (2) rating of categories of behavior directed toward adults (proximity seeking, contact maintaining, crying, etc.); and (3) a classification scheme based on profiles which combine several of the rated behavior categories. Results indicated that evidence for the stability of individual differences was a function of the level of analysis. The reliability of discrete behavior variables was typically very low, and there was little evidence of temporal stability. There was clear evidence for stable individual differences in the analysis of behavior category data. Classification data based largely on reunion behavior and crying were even more stable across the six month interval. These classifications have a variety of theoretically significant correlates in non-Strange Situation behavior from early infancy into the third year of life. (Author/SB)