The model of bulk-synchronous parallel (BSP) computation is an emerging paradigm of general-purpose parallel computing. In this paper, we consider the parallel complexity of graph colouring. A graph on n nodes with m edges and maximum degree Δ can always be properly coloured in Δ+1 colours. Gebremedhin et al. proposed a coarse-grained, distributed-memory algorithm for proper Δ+1-colouring that runs on a p-processor BSP-like computer in optimal O(m/p) local computation, optimal O(m/p) communication per processor, and O(p²) barrier synchronisation steps. We improve on their result by proposing a relatively simple, coarse-grained BSP algorithm for proper Δ+1-colouring, that achieves optimal local computation and communication, but requires only O((log p)²) barrier synchronisation steps.