Replication is the process of copying a database to another server, and then keeping the two copies in sync, so that they behave as near as possible.
Replication copies data or changes of data from one database to an another. Accessing the first database or the second, it does not matter because they are the same, i.e. they are in sync. Replication is not limited to two databases, any number of database can take place into the replication model.
It does not mean that the replicated databases are exactly the same, because replication can be configured that only a subset of tables or columns or rows will be replicated, i.e. partial replication. However, replication ensures that those specific configured objects (others are left alone) are kept in sync between the different databases.
With replication, maintenance of multiple copies of data to different locations is automated. Because each location has its one database, each location works essentially local with maximum performance and security; databases don't have to be accessed outside the local network or workstation.
Replication is a process which needs no interaction from the user. In that way replication is a low level process which is (or should be) completely invisible for the application user. This means that replication by copying databases (through backup & restore) or by using export- and import-scripts are not viable replication mechanisms.