We numerically study a directed small-world network consisting of attractively coupled, identical phase oscillators. While complete synchronization is always stable, it is not always reachable from random initial conditions. Depending on the shortcut density and on the asymmetry of the phase coupling function, there exists a regime of persistent chaotic dynamics. By increasing the density of shortcuts or decreasing the asymmetry of the phase coupling function, we observe a discontinuous transition in the ability of the system to synchronize. Using a control technique, we identify the bifurcation scenario of the order parameter. We also discuss the relation between dynamics and topology and remark on the similarity of the synchronization transition to directed percolation. © 2010 American Institute of Physics. ͓doi:10.1063/1.3476316͔The adjustment of phase and frequency in large systems of oscillatory units can lead to global coherent oscillations, i.e., synchronization. On the other hand, noise and heterogeneity in the system can weaken synchronization, or even destroy it. Synchronization in the nervous system can facilitate the transfer of information or cause epileptic seizures. Multistability and hysteresis of normal and pathological collective behavior are observed. When all oscillators are identical and the coupling tends to decrease phase differences a state of complete synchronization is asymptotically stable. But even in random networks with uncorrelated and homogeneously distributed node degrees this absorbing state may not be reached or disappear when it is perturbed locally. Here we perform a detailed numerical analysis of the transition between different states of synchronization in a directed smallworld network of phase oscillators. By varying the mean in-degree of the network or the nonlinearity of the phase coupling function at zero phase difference, we find discontinuous and continuous transitions with mean field critical behavior.