The invention provides a building method of a grid map by a mobile robot based on real-time data fusion. The method comprises the following steps: artificially dividing an environment into a plurality of grids with the same size, and obtaining the distance information by sonar sensors arranged on the front end of the mobile robot; extracting measured values of three sonar sensors which are closest to a currently computed grid unit at the same time, respectively explaining a single sonar data by a fuzzy logic and a probability theory so as to obtain a group of characteristic vectors which are subject to the data fusion as input vectors of a neural network, wherein, output values of the neural network comprises idle state, busy state and undefined state of the grid; finally updating the state of the grid by the Bayes rule. The mobile robot of the invention obtains the environmental information by sonar ranging finders and completes the environmental modeling, thus providing reliable basis for the subsequent autonomous navigation of the mobile robot.