As specified in RFC 790, each network class has a default subnet mask associated with it.
As shown in Figure 1, class A networks used the first octet to identify the network portion of the address. This is translated to a 255.0.0.0 classful subnet mask. Because only 7 bits were left in the first octet (remember, the first bit is always 0), this made 2 to the 7th power, or 128 networks. The actual number is 126 networks, because there are two reserved class A addresses (i.e., 0.0.0.0/8 and 127.0.0.0/8). With 24 bits in the host portion, each class A address had the potential for over 16 million individual host addresses.
As shown in Figure 2, class B networks used the first two octets to identify the network portion of the network address. With the first two bits already established as 1 and 0, 14 bits remained in the first two octets for assigning networks, which resulted in 16,384 class B network addresses. Because each class B network address contained 16 bits in the host portion, it controlled 65,534 addresses. (Recall that two addresses were reserved for the network and broadcast addresses.)
As shown in Figure 3, class C networks used the first three octets to identify the network portion of the network address. With the first three bits established as 1 and 1 and 0, 21 bits remained for assigning networks for over 2 million class C networks. But, each class C network only had 8 bits in the host portion, or 254 possible host addresses.
An advantage of assigning specific default subnet masks to each class is that it made routing update messages smaller. Classful routing protocols do not include the subnet mask information in their updates. The receiving router applies the default mask based on the value of the first octet which identifies the class.