REJECT vs DROP when using iptables
As a general rule, use REJECT when you want the other end to know the port is unreachable' use DROP for connections to hosts you don't want people to see.
Usually, all rules for connections inside your LAN should use REJECT. For the Internet, With the exception of ident on certain servers, connections from the Internet are usually DROPPED.
Using DROP makes the connection appear to be to an unoccupied IP address. Scanners may choose not to continue scanning addresses which appear unoccupied. Given that NAT can be used to redirect a connection on the firewall, the existence of a well known service does not necessarily indicate the existence of a server on an address.
Ident should be passed or rejected on any address providing SMTP service. However, use of Ident look-ups by SMTP serves has fallen out of use. There are chat protocols which also rely on a working ident service.
EDIT: When using DROP rules: - UDP packets will be dropped and the behavior will be the same as connecting to an unfirewalled port with no service. - TCP packets will return an ACK/RST which is the same response that an open port with no service on it will respond with. Some routers will respond with and ACK/RST on behalf of servers which are down.
When using REJECT rules an ICMP packet is sent indicating the port is unavailable.
The difference is that the REJECT target sends a reject response to the source, while the DROP target sends nothing.
This can be useful e.g. for the ident service. If you use REJECT then the clients doesn't need to wait for timeout.
More about this: http://www.linuxtopia.org/Linux_Firewall_iptables/x4550.html
Usually, you want to ignore probes from attackers to certain ports, by which I mean you do not want to send back 'connection refused'. 'Connection refused' means: 'there is a server here', and possibly gives away more information, whereas dropping a packet doesn't give away clues about software versions, possible vulnerabilities or even the fact that a server is listening at you IP.
The above is one of the main reasons to use DROP instead of REJECT.
I see lots of conflicting answers here and given this is the first article in Google with the right keywords; here is the correct explanation.
DROP does nothing at all with the packet. It does not get forwarded to a host, it does not get answered. The manpage of IPtables says it drops the packet on the floor, i.e. it does nothing with the packet.
REJECT differs to DROP that it does send a packet back, but the answer is as if a server is located on the IP, but does not have the port in a listening state. IPtables will sent a RST/ACK in case of TCP or with UDP an ICMP destination port unreachable.
If you're trying to hide your machine's existence entirely,
-j DROP is appropriate. For example, you might use this to implement a blacklist.
If you're trying to hide the fact that a port is open, you should mimic the behavior that would occur if the port was not open:
-p tcp -j REJECT --reject-with tcp-reset
-p udp -j REJECT --reject-with icmp-port-unreachable
If a port scanner sees that a few ports are dropping packets while most are rejecting them, it can assume the dropped packets are on ports that are open but hidden.