"funny" reassociating problem

[henrikg at fys.ku.dk]

Hi

I have the following problem for which I can't find an explanation. 
I've written a tool, which cycles through 3 APs, and collect statistics.
The problem is, that the associating process isn't stable. 
It can happen that I'm only able to connect to one AP at the beginning but then this is the only AP I can connect to at all, no matter where in the bulding I am located and how often I try to connect to the other APs. It seems arbitrary which AP I can connect to at first.
Trying to connect to XX:XX:XX:XX:XX.... failed
Trying to connect to YY:YY:YY:YY:YY.... success 
Trying to connect to ZZ:ZZ:ZZ:ZZ:ZZ:ZZ.... failed
 
Trying to connect to XX:XX:XX:XX:XX.... failed
Trying to connect to YY:YY:YY:YY:YY.... success 
Trying to connect to ZZ:ZZ:ZZ:ZZ:ZZ:ZZ.... failed
and so on
So in this case, reassociating to the YY:YY:YY:YY:YY AP is stable.

It also happens, that I can connect to all three APs in the beginning but then at some point I can't connect to one(two) of them, and this continues

Trying to connect to XX:XX:XX:XX:XX.... success
Trying to connect to YY:YY:YY:YY:YY.... success 
Trying to connect to ZZ:ZZ:ZZ:ZZ:ZZ:ZZ.... succes
and so on...

suddenly
Trying to connect to XX:XX:XX:XX:XX.... failed
Trying to connect to YY:YY:YY:YY:YY.... failed 
Trying to connect to ZZ:ZZ:ZZ:ZZ:ZZ:ZZ.... succes

Trying to connect to XX:XX:XX:XX:XX.... failed
Trying to connect to YY:YY:YY:YY:YY.... failed 
Trying to connect to ZZ:ZZ:ZZ:ZZ:ZZ:ZZ.... succes
and so on.
The connection to ZZ:ZZ:ZZ:ZZ:ZZ (in this case) stays stable.

First i thought I was out of range, but the same problem arises even standing right next to the APs. After trying to associate to an AP my prgramm sleeps for a second, and if it fails to connect it will try again 5 times. 
Can someone help me?