This paper deals with the discrete berth allocation problem with stochastic arrival times of vessels. The problem is to assign incoming vessels to a finite set of berthing points (berths) and to schedule them. The major objective is to minimize the expected total turnaround time of the vessels. We develop several new proactive, reactive and proactive/reactive approaches to this problem. Numerical experiments enable to compare these approaches. They show the impact of using full stochastic information instead of using mean values in a deterministic setting. The proactive/reactive approach, based on iterated tabu search and stochastic dynamic programming, provides good results when uncertainties on arrival times remain limited, while requiring only a few minutes of computing time on average. For larger levels of uncertainty, the proposed pure reactive approach clearly outperforms the others.