This paper addresses the central spanning tree problem (CTP). The problem consists in finding a spanning tree that minimizes the so-called robust deviation, i.e. deviation from a maximally distant tree. The distance between two trees is measured by means of the symmetric difference of their edge sets. The central tree problem is known to be NP-hard. We attack the problem with a hybrid heuristic consisting of: (1) a greedy construction heuristic to get a good initial solution and (2) fast local search improvement. We illustrate computationally efficiency of the proposed approach.