The train timetabling problem (TTP) consists of finding a feasible timetable for a number of trains which minimises some objective function, e.g., sum of running times or deviations from ideal departure times. One solution approach is to solve the dual problem of the TTP using so-called bundle methods. This paper presents a new bundle method that uses disaggregate data, as opposed to the standard bundle method which in a certain sense relies on aggregate data. We compare the disaggregate and aggregate methods on realistic train timetabling scenarios from the Iron Ore line in Northern Sweden. Numerical results indicate that the proposed disaggregate method reaches better solutions faster than the standard aggregate approach. © 2020 Elsevier Ltd