Topological maps represent the world as a network of nodes and arcs: the nodes are distinctive places in the environment and the arcs represent paths between places. A significant issue in building topological maps is defining distinctive places. Most previous work in topological mapping has concentrated on using sonar sensors to define distinctive places. However, sonar sensors are limited in range and angular resolution, which can make it difficult to distinguish between different distinctive places. Our approach combines a sonar-based definition of distinctive places with visual information. We use the robot' s sonar sensors to determine where to capture images and use cues extracted from those images to help perform place recognition. Information from these two sensing modalities is combined using a simple Bayesian network. Results described in this paper show that our robot is able to perform place recognition without having to move through a sequence of places, as is the case with most currently implemented systems.