Technical Abstract: The quality of rainfall-runoff simulation by Artificial Neural Networks (ANN) very much depends on the selection of appropriate input variables and design of the neural network. Here, the performance of three ANN designs that account differently for the effects of seasonal rainfall and runoff regime are investigated for their ability to simulate monthly rainfall-runoff on an 815 km2 watershed in central Oklahoma. Performance was measured in terms of coefficient of determination of the linear regression between observed versus simulated runoff, slope and intercept of the regression, Nash-Sutcliffe efficiency coefficient, and flow duration curves. The ANN design that accounted explicitly for seasonal regime of rainfall and runoff performed best by all performance measures. Explicit representation of seasonal regime was achieved by use of a separate ANN for each calendar month. While such a design limits the amount of training and test data, simulation results were superior in a month-by-month, year-round, and time series evaluation. Average coefficient of determination for the month-by-month evaluation was 0.83, for the year-round evaluation it was 0.87, and the Nash-Sutcliffe efficiency coefficient for the time series evaluation was 0.79 However, all three ANN designs displayed an average regression slope slightly under 1 and positive intercept, pointing to a tendency to under-predict high and over-predict low runoff values. A simple post-ANN runoff correction was proposed to reduce the noted under/over prediction tendency.