Two concurrent processes are said to be in conflict if they can get trapped in a situation where they both are waiting or running endlessly, forever unable to complete their common task. In the design of reactive systems, this is a common fault which can be very subtle and hard to detect. This paper studies conflicts in more detail and characterises the most general set of behaviours of a process which certainly leads to a conflict when accepted by another process running in parallel. It shows how this set of certain conflicts can be used to simplify the automatic detection of conflicts and thus the verification of reactive systems.