A common problem faced in robotic manipu- lation is developing techniques for accurate lo- calisation and mapping of 3D objects. Many techniques already exist to aid in estimating the structure of the world using information from a robot’s sensors such as stereo cameras, time-of-flight or structured light. These sen- sors and techniques used for modelling can of- ten be made accurate enough for most practi- cal applications (such as picking-up an object). However, some applications require a higher de- gree of accuracy (sub-millimeter) that is diffi- cult to achieve with the information available from these sensors. This paper proposes the use of tactile exploration to incrementally im- prove the accuracy of a prior 3D object model as the robot touches different parts of a work- piece. A modified Unscented Kalman Filter (UKF) has been developed to fuse the touch probe data with the existing model and refine it over time. The approach presented in this paper is intended for applications that require a high degree of accuracy and reliability (such as medical procedures) and as such, focuses on three primary requirements—accuracy, robust- ness and practicality.