@Stake published a paper (in two parts) on Bluetooth security. The author, Ollie Whitehouse, is also the developer of RedFang, a brute force attack tool exploiting known design characteristics (note that RedFang works on a range of MAC addresses, therefore the attacker has to know this information beforehand).
In this paper the author introduces the concept of "war-nibbling", very similar to war-driving, but applied to PAN (Personal Area Network) devices with short range.
Although most Bluetooth devices have a 10m range and communications are impacted by obstacles like doors and walls, new devices allow up to 100m range, creating conditions for anonymous connections (if in public, of course). In the first document the author explains how the security mechanisms implemented in the protocol work, and how users can set basic configuration items to prevent unwanted connections.
An interesting concept is the so called "Sweet-tooth", a Bluetooth Honeypot. Although still in development, the tool promisses some interesting features to trap eventual attackers to a Bluetooth enabled device.
Even with short range, Bluetooth attackers can use these communications to access services, like DUN (dial up) via a mobile phone, or piggyback into someone's else broadband network if there's a Bluetooth LAN Access Point open and vulnerable (some BT LAP have a non-PIN requirement as default setting).
If you're just interested in knowing more about this technology, or work with security in an environment where this is being deployed, it's an interesting reading.