In 1992, Bandelt and Dress introduced the split decomposition method that takes as input a distance matrix and produces as output a collection of splits that are not necessarily compatible. This method was the motivation for the SplitsTree program, and different versions have been written by Rainer Wetzel, Daniel Huson and David Bryant. In practice, split decomposition has turned out to be very conservative method and it becomes more and more timid as the size of the input set grows. More recent methods such as Neighbor-net, consensus networks, bootstrap networks or the Z-closure network are much more potent at producing splits and this puts new demands on algorithms for representing them. In this talk we will illustrate this problem and discuss some ideas for addressing it.