The NetBIOS datagram service is accessible through UDP port 138. As the NetBIOS name service is vulnerable to various naming attacks (resulting in denial of service in some cases), so can the NetBIOS datagram service be used to manipulate the target host and its NetBIOS services.
Anthony Osborne of PGP COVERT labs published an advisory in August 2000 that documented a NetBIOS name cache corruption attack that can be launched by sending crafted UDP datagrams to port 138. The full advisory is available at http://www.securityfocus.com/advisories/2556.
RFC 1002 defines the way in which Windows NetBIOS host information is encapsulated within the NetBIOS datagram header. When a browse frame request is received (on UDP port 138), Windows extracts the information from the datagram header and stores it in the NetBIOS name cache. In particular, the source NetBIOS name and IP address are blindly extracted from the datagram header and inserted into the cache.
A useful scenario in which to undertake this attack would be to send a crafted NetBIOS datagram to the target host, that mapped a known NetBIOS name on the internal network (such as a domain controller) to your IP address. When the target host attempts to connect to the server by its NetBIOS name, it instead connects to your IP address. An attacker can run SMBRelay or LC4 to capture rogue SMB password hashes in this scenario (which he can then crack and use to access other hosts).
Interestingly, Microsoft didn't release a patch for this issue: due to the unauthenticated nature of NetBIOS naming, it's a fundamental vulnerability! The MITRE CVE contains good background information within CVE-2000-1079.