What do we mean when we say that a group knows something? Reasoning about knowledge is of importance in modeling and analysing multi-agent systems and other distributed systems, and various notions of group knowledge have been proposed and shown to be useful. In my talk I will review these, argue that many of them are problematic, and propose some alternatives and extensions. The talk will be based on standard formalisations of knowledge and belief in modal logic. I will discuss a natural range of group belief concepts with distributed and general ("everybody-believes") as the two extreme endpoints and with many intermediate concepts in between. Distributed knowledge is sometimes described as what the members of the group would know of they "pool their knowledge together". This is inaccurate at best: for example, it is consistent that a group has distributed knowledge of a Moore sentence involving one of the members of the group (a sentence which cannot be known by that member, no matter how much "pooling" has taken place). I will also present and discuss a new group modality that actually captures what is true after the group have fully shared their information with each other -- after their distributed knowledge has been resolved.