All IP addresses have a network and host portion. In classful addressing, the network portion ends on one of the separating dots in the address (on an octet boundary). Classless addressing uses a variable number of bits for the network and host portions of the address.
|<-------- 28 bits Network ------->||4 bits host|
Classful addressing divides an IP address into the Network and Host portions along octet boundaries. Classless addressing treats the IP address as a 32 bit stream of ones and zeroes, where the boundary between network and host portions can fall anywhere between bit 0 and bit 31. The network portion of an IP address is determined by how many 1's are in the subnet mask. Again, this can be a variable number of bits, and although it can fall on an octet boundary, it does not necessarilly need to. A subnet mask is used locally on each host connected to a network, and masks are never carried in IPv4 datagrams. All hosts on the same network are configured with the same mask, and share the same pattern of network bits. The host portion of each host's IP address will be unique.