This paper presents a back-propagation neural network method to estimate traffic flow volumes on signalized intersections. Based on volume estimates, an optimum cycle is then determined. We take five-minute intervals as term basis for volume estimation. Al-RABIAH intersection is used as a case study. Actual traffic flow data for the four main streets of the intersection have been collected for four weeks on a five-minute basis. Two weeks of the data were used for training the network. The other two weeks of the data were used for testing the network. Results obtained for various network structures are discussed. The approach is demonstrated to provide satisfactory results for practical use.