Reconstructing bomb trajectories resulting from Strombolian activity can provide insights into near-surface dynamics of the conduit system. Typically, the high number of bombs involved represents a challenge for both automatic and manual bomb identification and tracking methods. Here, we present a method for the automated recognition of hundreds of bombs (100 to 400 depending on the explosion observed) and for the reconstruction of their trajectories in time and 3D space by stereophotogrammetry. The data involve video collected at 30 Hz with two synchronised cameras (Basler 1300-30), separated by 11°, targeting explosions at Stromboli (Aeolian Islands, Italy) in September–October 2012. In total, six data sets were collected for emissions lasting less than 15 s. The 3D reconstructions provided more accurate velocity estimations (error < 10%) than 2D analyses (errors up to 90–100% for bombs moving parallel to the line of sight of the camera). By coupling the measured trajectories with a numerical ballistic model, we show that the method can be used to estimate the directional distribution of bombs and their velocities at the vent (which in this case was 30–130 m s−1), the wind velocity (~ 3.5 m s−1 from the NW) and the drag coefficients (10−3.5 − 10−0.5) of the bombs. The 3D reconstructions also provide a quantification of the directions of explosions and show that explosions can be radial, oriented in a predominant direction of ejection or in several directions; these dispersion patterns can change during a few seconds in a single explosion. We relate the changing directions of ejections to rheological variations in the upper part of the magmatic system probably filled with a mixture of partially crystallised magma which can direct rising slugs along preferential paths.