In this paper, we propose a robust ROI-based technique for needle detection and tracking in 3D ultrasound. Based on a previously proposed RANSAC procedure, a region of interest (ROI) selection strategy is implemented along with a Kalman method. Due to the scatter noise of ultrasound, the background in 3D ultrasound volumes usually features large echoes corresponding not only to the needle to be detected, but also to other structures, for example interfaces between tissues. Even if the RANSAC algorithm should handle situations with many outliers very efficiently, it still fails in some situations. A well-selected ROI strategy can lead to simpler situations by reducing the number of outliers, thereby increasing the so-called inlier ratio. The result is improved robustness of the RANSAC procedure. The simulation results visibly improve the proposed method compared to the existing RANSAC based tool localization method.