Replicated data can be propagated in a number of ways.
All propagation methods are based on either push subscriptions or pull subscriptions.
A subscriber can support a mixture of push and pull subscriptions simultaneously.
* Push Subscriptions
If you set up a push subscription,
the distributor is responsible for providing updates to the subscribers.
Updates are initiated without any request from the subscriber.
A push subscription is useful when centralized administration is desired
because the distributor, rather than multiple subscribers, controls and administers replication.
In other words, the initiation and the scheduling of the replication are handled on the distributor.
Push subscriptions allow you a lot of flexibility in scheduling replication.
Push subscriptions can be configured to propagate changes immediately
after they are made or to perform updates on a regular schedule.
* Pull Subscriptions
Pull subscriptions allow subscribers to initiate replication.
Replication can be initiated either via a scheduled task or manually.
Pull subscriptions are useful if you have a large number of subscribers
and if the subscribers are not always attached to the network.
Because subscribers initiate pull subscriptions,
subscribers not always connected to the network can periodically connect
and request replication data.
This can also be useful in reducing the number of connection errors reported on the distributor.
If the distributor tries to initiate replication to a subscriber that does not respond,
an error will be reported.
Thus, if the replication is initiated on the subscriber only when it is attached,
no errors will be reported.
Source: Microsoft SQL Server 2000 Administrator's Companion Book