Cross-sectional studies of personal injury accidents at samples of four-arm roundabouts, four-arm traffic signals and major/minor junctions on dual carriageways, and four-arm traffic signals on urban single carriageways are described. The sample of roundabouts includes separate groups of small island roundabouts, conventional roundabouts with mostly single carriageway approaches and roundabouts of conventional design on dual carriageways in both the 30-40 and 50-70 mile/hr speed limit zones. The dual carriageway traffic signals sample has separate groups for junctions with and without a separate stage for traffic turning right off the dual carriageway, also in both speed limit zones. Crossroads and right/left staggered junctions on 50-70 mile/hr dual carriageways comprise the major/minor junction sample. The single carriageway traffic signal junctions were all on 30 mile/hr roads; this sample was structured according to whether the junction (i) was operating under Urban Traffic Control (UTC) or not (ii) had a pedestrian stage or not and (iii) had two vehicle stages or more than two. Tabulations are given showing accident frequencies, severities and rates (per vehicle inflow) by junction category and of accidents by type (e.g. entering-circulating, approaching, right angle, etc.), road user involvement (e.g. pedal cyclist, motor cyclist, car driver, pedestrian, etc.) and other characteristics. A methodology for the analysis of accident frequencies using generalised linear modelling techniques is described. Predictive models are developed which relate accident frequencies, disaggregated by accident type and junction arm, to functions of the traffic and pedestrian flows and the geometric and control features of the junctions. The models quantify the effects of those features of design and operation which are important with respect to junction safety and provide an essential tool for traffic management appraisal. In particular the models show that at roundabouts entry path curvature is one of the most important determinants of safety; the interaction with entry width is also critical and suggests that for safety wide entries should only be used with substantial deflection. At the traffic signal junctions the provision of a separate stage for right turning traffic gives a substantial reduction in the rate of principal right turn accidents, while at the dual carriageway major/minor junctions right/left junctions are clearly shown to be safer than crossroads.