A random network model is used to generate the oxygen–oxygen pair correlation function of liquid water and amorphous solid water. The model is basically an extension of the bent hydrogen bond model proposed by Pople in 1951; it differs from the earlier work in that instead of being a parametric fit to the data, it predicts the dispersion in OO distances for each shell of neighbors, and the dispersion in OOO angles, from independent spectroscopic data. Our random network model is based on the hypothesis that there exists a separation of time scales for various classes of molecular motion in liquid water. In particular we assert that it is useful to regard libration and hindered translation as occurring in a quasistatic network of bonded molecules and diffusion and related relaxation processes as occurring much more slowly. The predicted OO pair correlation function is in good but not perfect agreement with experiment over a wide temperature range. As a further test the model is used to predict the dielectric constant of liquid water. Excellent agreement is found over the temperature range 10–300 °C. The relationship between the random network model and computer simulations of liquid water is extensively discussed.