This study investigates the onset phase of a strong Adriatic bora windstorm that occurred on 4 April 2002. The target area is a gap about 20 km wide embedded in the coastal mountain barrier of the Dinaric Alps that favours strong jet-like winds. Airborne-aerosol back-scatter lidar measurements on board the DLR Falcon research aircraft, together with surface and upper-air observations, are used to verify high-resolution numerical experiments conducted with the mesoscale atmospheric model RAMS and a single-layer shallow-water model (SWM). Especially during the breakthrough phase of the bora, the flow at the gap exit exhibits a complex spatial structure and temporal evolution. On a transect through the centre of the gap, a hydraulic jump forms; this is located close to the coast throughout the night, and starts to propagate downstream in the early morning. On a transect through the edge of the gap, a lee-wave-induced rotor becomes established, due to boundary-layer separation. It starts to propagate downstream about two hours after the jump. This flow evolution implies that the onset of strong winds at the coast occurs several hours earlier downstream of the centre of the gap than downwind of the edge of the gap. Consequently, the wind field in the vicinity of Rijeka airport, located downwind of the gap, is strongly inhomogeneous and transient, and represents a potential hazard to aviation. Measured bora winds at the surface exceed 20 ms −1 , and the simulated wind speed in the gap wind layer exceeds 30 ms −1 . The simulated turbulent kinetic energy exceeds 10 m 2 s −2 .RAMS indicates that wave-breaking near a critical level is the dominant mechanism for the generation of the windstorm. Gap jets can be identified downstream of several mountain passes. The simulated wave pattern above the Dinaric Alps, the wave decay with height due to directional wind shear and the strong flow descent on the leeward side of the barrier are supported by measured back-scatter intensities. Basic bora flow features, including gap jets and jumps, are remarkably well reproduced by SWM simulations. The RAMS reference run captures observed flow phenomena and the temporal flow evolution qualitatively well. A cold low-level bias, an overestimated bora inversion strength, and a slightly too-early bora onset are probably related to insufficient turbulent mixing in the boundary layer. The amplitude of trapped gravity waves, the time of the bora breakthrough and the inversion strength are found to be quite sensitive to the turbulence parametrization.