You actually have several choices in connecting to both ISPs.
You can run BGP and load balance (sort of) with full resilience. This is the best way if you are providing externally accessed services, as it is the only way you can serve up the same IP address through both ISPs. Load sharing with BGP is possible, but takes work and is rarely exactly even.
The other option is to use policy routing to split your traffic between the two ISPs, as suggested by mkmodl. This provides load balancing (to the extent you can evenly divide traffic by assigning IPs from the appropriate ISP to internal systems. This option may or may not allow you to provide resilience to link failure (and does not provide resilience to ISP failure).
If you can reliably detect link failure at the link level (such as by using keep alives on an point-to-point T1), you can use NAT to make your inside users on the failed link's IPs appear as if they are in the remaining ISPs address space. This is often satisfactory for clients surfing the net, but is rarely adequate for web servers or other services you are providing to the outside world (other than SMTP, which can automatically handle multiple servers).
There is a brief white paper on my web site which goes over all the options for adding resilience to an ISP connection as well as extensive discussion and example configurations (including both BGP and NAT approaches) in chapter 8 of my book High Availability Networking with Cisco.
Good luck and have fun!
Vincent C Jones
www.networkingunlimited.com
