In this paper, we derive a topological pattern of urban street networks using a large sample (the largest so far to the best of our knowledge) of 40 US cities and a few more from elsewhere of different sizes. It is found that all the topologies of urban street networks based on street-street intersection demonstrate a small world structure, and a scale-free property for both street length and connectivity degree. More specifically, for any street network, about 80% of its streets have length or degrees less than its average value, while 20% of streets have length or degrees greater than the average. Out of the 20%, there are less than 1 % of streets which can form a backbone of the street network. Based on the finding, we conjecture that the 20% streets account for 80% of traffic flow, and the I% streets constitute a cognitive map of the urban street network. We illustrate further a peculiarity about the scale-free property.