For Quicksort and Mergesort we show empirically the strong in uence of branch mispredictions on the running time. This is in line with recent ndings of Sanders and Winkel [18], who demonstrate the practical importance of avoiding branch mispredictions in the design of sorting algorithms for current CPU architectures. For Heapsort, our experiments indicate that data cache misses are the dominant factor for the running time.