Previous research in bystander intervention found that the presence of other bystanders reduces helping behaviour in an emergency (bystander effect). This research was mainly conducted in the context of non-dangerous, non-violent emergencies. We hypothesize that the classic bystander effect does not occur in more dangerous situations because: a) they are faster and more clearly recognized as emergency situations; and b) higher costs for refusing help increase the accepted costs for helping. Following this line of reasoning, the present research tests whether the bystander effect is affected by the degree of the emergency's potential danger. Results supported our expectations: In situations with low potential danger, more help was given in the solitary condition than in the bystander condition. However, in situations with high potential danger, participants confronted with an emergency alone or in the presence of another bystander were similarly likely to help the victim.