Many computer vision applications require synchronization between image acquisition and external trigger events. Hardware and software triggering are widely used but have several limitations. Soft synchronization is investigated, which operates by time tagging both trigger events and images in a video stream and selecting the image corresponding to each trigger event. A stochastic model is developed for soft synchronization; and, based on the model, the uncertainty interval and confidence for correct image selection are determined, and an efficient calibration method is derived.Soft synchronization is experimentally demonstrated on a Linux image processing computer with a camera connected by the IEEE-1394 serial bus. To minimize timing variability, time tags are associated with images and events in the corresponding interrupt service routines within the operating system of the image processing computer. The model-based analysis applied to the experimental hardware shows that image/event synchronization within 1 2 inter-frame interval can be achieved with 99% confidence. Experiments confirm this result.