This paper presents an algorithm for fusing scans from a time-of-flight laser and odometer readings from the robot. The range weighted Hough transform is used as a robust method to extract lines from the range data. The resulting peaks are used as feature coordinates when these lines/walls are used as landmarks during navigation. The associations between observations over the time sequence are made in a systematic way using a decision directed classifier. Natural geometrical landmarks are described in the robot frame together with a covariance matrix representing the spatial uncertainty. The map is thus built incrementally as the robot moves. If the map is given in advance the robot can find its location and navigate relative to the map. Experimental results and simulations are presented for a mobile robot with a scanning range measuring laser with 2 cm resolution