Background: The diagnosis of attention-deficit/hyperactivity disorder (ADHD) in adolescence and adulthood is particularly challenging because retrospective confirmation of previous childhood ADHD is mandatory. Therefore, collecting valid diagnostic information about behavior at school is important. Primary school reports often contain descriptions of academic performance and social behaviors associated with ADHD criteria. Yet, there is no systematic approach available how to assess such reports quantitatively, and therefore, there is also no study on how valid such an approach could predict an ADHD diagnosis. Methods: We examined primary school reports from Germany (ADHD: n = 1197, typically developing controls: n = 656) for semantic references to ICD-10/DSM-5 main and sub-criteria of ADHD. Descriptions were assessed on a quantitative scale (blinded clinical expert rating) for disorder-associated behaviors (symptoms scale) as well as for desired, adaptive behaviors (competencies scale) according to these criteria. The scores of these developed scales have been summarized to summary scores. Scores were analyzed using linear mixed models, and sensitivity and specificity were estimated using receiver operating characteristics (ROC). Results: Ratings showed highly significant differences between school reports of children with and without ADHD. For the summary scores, both symptoms and competencies scales showed high diagnostic accuracy (ROC area under the curve at least 0.96) with best discrimination when combining both into an integrated index (sensitivity and specificity > 0.97). Conclusions: Our findings suggest that systematic quantitative analysis of primary school reports should be further explored to construct a valid instrument for retrospective assessment of childhood ADHD criteria to aid the diagnostic process in adolescents and adults.