Background Attachment representations of late‐adopted children have usually been measured by attachment narratives or observational procedures. Recently an attachment‐based coding system for family drawings was developed by attachment researchers and it was used both with clinical and nonclinical samples, but it has never been used with adoptees. Method This study examined the differences between attachment representations of 29 late‐adopted children aged 5–7 years (M = 6.35, 51.7% girls) and 12 non‐adopted peers as assessed by family drawings, controlling for demographic variables and children's cognitive status. The attachment‐based coding system of family drawings included three levels: (1) 24 individual markers, (2) eight global rating scales (1–7 points), and (3) four attachment categories (secure, avoidant, resistant, and disorganized). Results Late‐adopted children assessed with the family drawings were more insecure on the attachment categories and achieved lower scores on positive global ratings such as the Vitality/Creativity and Family Pride/Happiness scales, higher scores on the Role Reversal scale, and a tendency toward higher scores on the Bizarreness/Dissociation scale. No difference emerged between the two groups regarding the individual markers. Conclusions Family drawing seemed to be a useful tool for classifying attachment representations, and able to capture underlying mental states that it was hard for late‐adopted children to express in words.