This paper aims to study the predictive validity of mothers' attachment Q‐sort descriptions (AQS) for assessing attachment security in children in Japan. Attachment theory predicts that attachment quality will be related to quality of interaction with mother and peers. Forty‐eight mothers of two‐year‐old children described their children using the AQS. In a play session with another mother–child dyad, child–mother interaction and peer interaction were observed. Neither of the predicted relations of attachment security, i.e., to child–mother interaction or to peer interaction, were found. It is argued that these findings indicate lack of validity of the mother AQS as method for the assessment of attachment security.