With split tunneling, your VPN client is only able to send outgoing data through the VPN tunnel. This does offer some protection against intrusion -- for example, if an intruder tries to connect to a service running on your client, your client's responses will be forwarded over the VPN tunnel to the VPN gateway instead of being returned "in the clear" to the intruder. But what ultimately happens to those responses depends on the VPN gateway's configuration -- if the VPN gateway is configured to relay non-private traffic to the Internet, then the responses might be received by the intruder anyway.
Some VPN clients can be configured to drop incoming packets received outside a defined VPN tunnel. In this case, your client machine may ignore packets arriving from any source other than your VPN gateway WHEN THE TUNNEL IS ACTIVE. Many VPN clients go beyond this by integrating personal firewall software that blocks incoming connections received at any time. Some VPN clients have even been integrated with "scan on connect" features that check your client for infection before other traffic can flow through the VPN tunnel. In short, exactly what level of protection you have depends on your VPN client -- VPN tunnels per se don't protect your client machine, but related endpoint security measures and correctly-configured rules can detect and prevent client-side intrusion.
This was first published in July 2004