Due to the poor quality of ultrasound images, fully automatic segmentation methods are not feasible. This article describes a novel approach to the semiautomatic segmentation of ultrasound images.Although user interaction is not required much, it is used as an important factor and incorporated into the traditional B-spline snake models. The modified model is called an interactive B-snake model because the movement of the active contour is constrained through user interaction. By introducing a set of moving rules, B spline segments are moved to the desired boundary directly. The statistical models are trained on-the-fly by observing boundaries accepted by the user. The resulting algorithm is especially useful when dealing with successive slices and provides fast, reliable and verifiable segmentation in ultrasound images. The algorithm has been used successfully on the Liver Tumor Surgical Simulation System.