A trunk link can carry multiple vlans. So lets say on each switch you have vlans 5,6 & 7. If sw1 sends a frame for vlan 6 to sw2 how does sw2 know which vlan that packet is in ? It knows because sw1 adds a vlan tag to the frame header and sw2 reads this vlan tag and sees that the packet is in vlan 6.
The native vlan is the vlan that is not tagged. So lets say in the above example vlan 5 is the native vlan. When sw1 sends a packet for vlan 5 there is no vlan tag added to the frame. So when sw2 receives the frame there is no vlan tag to read. So it must agree on the native vlan otherwise it wouldn't know which vlan the frame is meant to be in. If sw2 had the native vlan set to 7 for example then it would think the frame it just received was in vlan 7. This would be bad because now you have just "joined" two vlans together.
If you do configure different native vlans on either end of the trunk and you are running CDP then you would get CDP Native vlan mismatch errors in your log.
Note that the native vlan must only agree per trunk link so you could if you wanted have different native vlans per trunk link as long as each trunk link agreed on either end but in practice you generally use the same native vlan across all trunk links.
Finally, by default the native vlan is vlan 1 on all Cisco switches. It is recommended to change this to another vlan. The native vlan should have no ports assigned to it and it does not need an SVI because there is never a need to route native vlan.
Jon
