This paper presents a new method of real-time scene recognition for robots, which utilizes a two-dimensional (2-D) gray-scale image of three-dimensional (3-D) objects in the scene, and a genetic algorithm (GA).In the scene recognition process, a model shaping a target object is employed as the knowledge base for recognition purposes. GA is employed for optimization of the evaluation function that expresses the degree of matching between a target beeing imaged and a model shaping the target. Here, the optimization means simultaneous recognition of a target and detection of its position/orientation. Also we are interested here in using GA, since it strikes a good balance between exploration of the search area and exploitation of the results obtained by this search process, without need to scan whole the 2-D image plane. Hence, GA will provide faster recognition performance to the vision system. This GA-based scene recognition method can be designated as "evolutionary scene recognition method", since for every step of the GA's evolution, it struggle to perform the recognition of a target in the new input image of the recognition system. To evaluate the effectiveness of the proposed scene recognition method, experiments are performed and some results are presented.