We examined the stability of ratings on the Hazan and Shaver (1987) single-item attachment style scale in a number of data sets, gathered by us and other researchers. Approximately 30% of subjects overall changed their attachment style classifications over a relatively short time span (ranging from 1 week to several months). The highest rate of instability was observed in subjects who classified themselves as anxious-ambivalent–the majority of whom changed their ratings from one time to the next. Given these findings, we explore the methodological and conceptual implications of instability in attachment style ratings. With regard to the former, we question the current practice of selecting subjects for participation in research based on responses to the attachment style questionnaire administered on a different occasion. Our findings suggest that a substantial proportion would change their style rating in the interim. In terms of conceptualization, we examine a number of different explanations for the observed instability and propose that it may reflect variability in the underlying construct, rather than a lack of continuity in style or unreliability of measurement. From this perspective, an individual's response to an attachment style questionnaire reflects the relational schema that is activated at that moment, rather than an enduring general disposition or trait. Stability in ratings is therefore neither assumed nor expected.