As you may have noticed when configuring your AP, there are 11 channels available in the U.S. for operation in the 2.4 GHz band. (The 2.4 GHz band is used by 802.11b, 802.11g and draft 802.11n gear.)
Figure 2 provides a visual representation of the channels and frequencies.
Figure 2: Depiction of 2.4GHz frequencies for 802.11b/g channels.
(Cisco Press, 2006) by Eric Geier , used by permission
While technically you can choose any of these channels, in practice, you should use only channels 1, 6, or 11. The reason for this limitation is channel overlap. 802.11b and g use eleven channels in the 2.4GHz band, spaced at 5MHz intervals. Since the commonly accepted width of each channel is 22MHz for 802.11b and 20MHz for 802.11g, both 802.11b and g are said to have three non-overlapping channels (1, 6 and 11).
Tip: Now-defunct wireless networking management company Cirond argued that there are actually four channels (1, 4, 8, 11) that can be used for 802.11b and g with virtually no performance penalty.
Now if all the energy of the transmitted signal actually were contained within a 20 (or 22MHz) band, the definition of "non-overlapping" might be simpler. But reality is somewhat more complicated.
Figure 3: 802.11b Transmit Spectrum Mask
From Matthew Gast's 802.11 Wireless Networks: The Definitive Guide , used by permission
Figure 3 shows an idealized spectral plot (power vs frequency) of an 802.11b signal. To paraphrase the explanation in Chapter 10 of Matthew Gast's excellent book, this plot shows that transmitted power is reduced by 30dB (1/1,000) below the power at the center of the channel (that's what the dBr notation means) at +/-11MHz away from the channel center and 50dB (1/100,000) below at +/-22MHz away.
NOTE: The following spectrum diagrams are based on Figure 3 and are not done to exact scale. Any inaccuracies are not intentional!
Since 11b and g channels are on 5MHz spacings, two channels right next to each other (1 and 2 for example) would overlap as shown in Figure 4.
Figure 4: 802.11b adjacent channel overlap
The yellow shaded area represents the power from channel 2's signal that overlaps into channel 1's main lobe (the largest "hump" and also the frequency band that contains most of the signal's power). Since a significant amount of channel 2's main lobe overlaps into channel 1's main lobe (and vice versa), communication on both channels will suffer. Contrast this picture with the situation shown in Figure 5.
Figure 5: 802.11b "non-overlapping" channel overlap
This figure has the same scale as Figure 4, but shows signals in the "non-overlapping" channels 1, 6 and 11. Since the power from each signal doesn't magically stop at the 22MHz channel boundaries, there is still overlap between "non-overlapping" channels. But in this case, the yellow shaded area that represents channel 11's power that is overlapping into the main lobe of channel 6 is at least 30 dB lower (1/1000) than channel 11's peak power.
Put simply, channels 1, 6 and 11 are considered to be "non-overlapping" because the amount of power that does overlap is supposedly too small to significantly affect each channel's operation. Whether that's actually the case, however, depends on many other factors, including the device's Adjacent Channel Rejection (ACR) capability, and, of course, the physical distance between devices on different channels. By the way, although I've been using examples based on 802.11b, the situation is pretty much the same for 802.11g.
Note: For a better picture of what real 802.11g signals look like when captured on a spectrum analyzer you can refer to some screen shots shown here.