I've read several messages in this forum section about "implementing a Kalman filter". I would like to do exactly that to reduce the uncertainty contained in my ultrasonic data but I do not know how this can be done with Arduino. Does someone knows if there is a nice library somewhere which can be used to do that?

Look at http://en.wikipedia.org/wiki/Kalman_filter

You will see it is a recursive filter, that is the output depends on the input and previous states of the output.

This is a digital filter, the arduino is not well suited to digital filters due to:- 1) the lack of memory for a buffer 2) the length of time it takes to do floating point multiplication 3) the comparative low speed of the A/D converter 4) the low resolution of the A/D converter

You can make one but it is not very good. It is very similar to the filters you see here:- http://interface.khm.de/index.php/labor/experimente/arduino-realtime-audio-processing/

Thanks a lot for your answer!
I realize Arduino cannot handle this type of heavy calculations. In fact what I had in mind was to transfer the ultrasonic data from Arduino to Processing and do the statistics there. I thought people on this forum might be more interested in data treatment than Processing people.
If you have any clue who might be the more knowledgable people about kalman filter implementation for gyroscope or accelerometer data, let me know.

Use fixed point or integer arithmetic. No-one in their right mind would try and use FP maths on something like an AVR for fast maths