A before-and-after study was carried out of injury accidents involving bicyclists on 91 roundabouts in Flanders-Belgium. The study design accounted for the effects of general safety trends and regression-to-the-mean, but could not take into account the possibility of specific changes in traffic volume at roundabouts. The conversion of intersections into roundabouts produces a significant 27% increase in the number of injury accidents involving bicyclists on or nearby the roundabouts. The increase is even higher for accidents involving fatal or serious injuries (41-46%). Compared to the proven favourable effects of roundabouts on safety in general, this result is unexpectedly poor. However, the effects of roundabouts on bicycle accidents differ depending on whether these roundabouts are built inside or outside built-up areas. When inside built-up areas, the construction of roundabouts increased the number of injury accidents involving bicyclists by 48%. For accidents causing fatal or serious injuries inside built-up areas, an average increase of 77% was found. However, outside built-up areas the zero-hypothesis of 'no safety effect for bicyclists' cannot be rejected (best estimate: +1% accidents, not significant). Roundabouts that replace traffic signals perform worse when compared to roundabouts at other types of intersections.