Which K-factor should be used in case players have different K-factor values

I am implementing a ranking system based on the Elo rating and cannot get a point about the K-factor. If two players with different skills and, therefore, different K-factors are playing, which exactly K-factor should be used when changing their ratings?

For example, player A has rating 2500 and K-factor 16 (probability ~75%) while player B has rating 2300 and K-factor 24 (probability ~25%). If player A wins, do I need to use 16 as K-factor for both players or 16 for player A and 24 for player B?

-
From what I understand it's unlikely that a player with a lower rating would have a higher K-factor than a player with a higher rating. Typically the higher a player's rating, the lower their K-factor. – Byte56 Jun 1 '12 at 18:12

Each player's new ranking is calculated based on their own K factor. It is just a scaling value to ensure that values converge at a reasonable rate, and can be arbitrarily chosen according to whatever suits your game. In particular, if a new player plays against an old player you usually want a large K factor for the new player to make their rating change more significantly, whereas the older player will have a lower K factor since it's unlikely their rating is a significant over or underestimate of their skill.

http://en.wikipedia.org/wiki/Elo_rating_system#Most_accurate_K-factor

-
This is one way to do it. I believe it's also common to have a single K factor for the game. – Byte56 Jun 1 '12 at 18:54