The use of infrared range-finder sensors in the environmental recognition system for mobile robot has the advantage of low sensing cost compared with the use of other vision sensors, such as laser range-finder and CCD camera. This paper presents a methodology building the high accuracy environmental map using a mobile robot with low cost infrared range-finder sensors. In the map building using an infrared range-finder sensor, the errors due to non-linearity, specular reflection are contained. In this paper, therefore, the error due to non-linearity is compensated using a neural network. The neural network used consists of multi-layer perceptron and Levenberg-Marquardt algorithm is applied to learn it. And also, the random error of readings and the uncertainty of environment are taken into sensor modeling at probabilistic approach. The map is represented by occupancy grid framework and updated by the Bayesian estimation mechanism. The effectiveness of the proposed method is verified through experiments.