In the last years the one-sided communication paradigm has become important in the programming of distributed memory machines. PGAS-like APIs, and also the one- side communication facilities of MPI, have evolved significantly and now open up new opportunities for the development of HPC applications. In this paper we present a model covering the essential features of one-sided communi- cation systems for discussing and comparing their operational semantics. Our approach is based on task graphs, which we have extended by introducing virtual tasks. A virtual task represents an asynchronous communication operation performed by the underlying system or a RMA unit. By this means we describe the fundamental functions of three popular one-sided communication APIs, namely OpenShmem, MPI 3.0, and GASPI. We conclude the paper with an outline of a parallel algorithm based upon our model, which can be used for data race detection and performance tuning.