Abstract: Data packets (D1, D2) each assigned to one of at least two priority classes (P1, P2) are stored in a buffer (MEM2) designed to hold data on a threshold value for each priority class wherein, instead of having the threshold fixed for each priority class whereby the data packets of the corresponding priority class will no longer be written into the buffer (MEM2) when the threshold is exceeded ("partial buffer sharing"), one or more thresholds are dynamically controlled. The current loss probability of the highest priority class (P1) is determined, and the threshold (MEM1) of the lowest priority class (P2) is then raised or lowered (REG1). This results in improved buffer utilization in applications which require a given loss probability for high priority classes, and calculation of a threshold value is no longer necessary.