One of the significant problems facing the water resource engineer is calculating the coefficient of roughness for subsequent design calculations of the discharge amount of a channel or river. In this study, experiments were conducted in a semi-circular, straight channel to investigate the factors affecting bed roughness and flow discharge using Artificial Neural Network (ANN). For this purpose, three semi-circular channel models with free overfall were constructed and installed in a 6-meter-long laboratory flume. The length of these models was 2.50 m with three different diameters (D= 150, 187, and 237mm) and three bed slopes (S=0.004, 0.008, and 0.012). Three sand particle sizes (ds) were used for each semi-circular channel to roughen the bed. The results showed that the Manning roughness coefficient obtained using a rough bed surface was higher than the channel with a smooth bed surface. Also, the results revealed that the Manning roughness coefficient and the Froude number were inversely related. (ANN) analysis showed a good agreement between the experimental and predicted results of flow and roughness. The bring depth (yb) had an 85.8% impact percentage on the free overfall discharge for semi-circular channels, while the bottom slope (S) had only 1.1%.