Reicher has recently developed the social identity model of crowd behaviour based on self-categorization theory (SCT). This model begins to tackle the thorny theoretical problems posed by the dynamic nature of crowd action (Reicher, 1996b). The present paper describes an ethnographic study of a crowd event in which there were changes in the inter-group relationships over time. It is suggested that the laboratory evidence in support of SCT is complemented by ethnographic research of this type. By exploring situations in which definitions of context and/or categories are not purposefully manipulated, we can demonstrate the explanatory power of a dynamic and interactive approach to social categorization.