With a large number of surveillance cameras, it is not an easy task to determine which camera should be monitored and which region of the camera images should be checked so that all the activities and/or events in a scene are examined. We present a new method to realize effective visual surveillance under an environment in which a number of noncalibrated fixed surveillance cameras are being operated. We also show two applications that are useful for surveillance tasks based on our proposed method. One is “suggestion of associative blocks”, and the other is “dominant camera selection”. Our approach exploits co-occurrence between two regions of interest (ROIs) over the surveillance cameras, and it needs neither calibration nor supervised training. We have conducted preliminary tests with forty cameras installed in a room and a corridor nextto the room, and some promising results of the two applications are shown in this paper.