This study aimed to develop a new scale to examine primary and secondary school students' perceptions of the severity of cyberbullying behaviours, and to explore further whether differences exist in the means of gender, grade and participant role. A total of 707 primary and secondary school students (M = 14.7) in Taiwan participated in this study. Two Olweus-like global items were used to identify students' participant roles. A self-reported cyberbullying severity scale (CSS) was developed and validated by Rasch measurement. Results of this study supported the reliability and validity of the 16-item CSS. Impersonation was rated as the most serious type of cyberbullying. Cyberbullying behaviours that occurred in private were rated as less severe than were those that occurred in public. A Rasch latent regression analysis revealed that some gender and involvement effects were found, but no statistically significant difference was found among means of four participant roles. The behavioural hierarchy of cyberbullying severity, mean differences among personal attributions and cyberbullying intervention are discussed at the end of the article.