Numerical codes are being used to simulate flood progress in cities but no real knowledge about the accuracy of these numerical calculations are available. The objective of this study is to check the capabilities of a 2D code to model a flood in an urban area. A simplified city flood was then modelled experimentally and its data were compared with the simulation results. Although results can be considered accurate enough for operational purposes, water depth differences remain, even with a refined mesh. In order to find out which typical urban situations are difficult to simulate by the code, two experimental test cases representing urban details were then modelled and calculated. The first one is the flow around a building where the reflections of the wave on the walls and recirculation behind the building are of major importance. The second one is the distribution of flow discharge in a four branches crossroad with two incoming and two outgoing streets with the same width and same bottom slope of 1%.