This paper develops a theory of sound generation by orbital motion of sea surface waves (as distinct from motion directly resulting from wave breaking such as bubble oscillation) and compares the noise predictions with measurements in a carefully controlled experiment. Theory and measurement were found to agree within the experimental errors. The mechanism is also known as the nonlinear interaction of surface waves and has been addressed by a number of authors. The approach of this paper differs from other models in that it avoids the use of the commonly applied perturbation expansion, and calculates the total noise field whereas others have limited their estimates to sound production from standing waves and waves that closely approximate standing waves. It is shown that while standing waves result in distributed dipoles with vertical axes, other wave interactions result in dipoles with axes inclined to the vertical so that there are components with both horizontal and vertical axes. The relative contribution of the horizontal dipole components to the noise field is of the same order of magnitude as that of the vertical dipole components. This paper therefore predicts higher noise levels and different directionalities, and also determines the contribution from the evanescent or near field that dominates for receiver depths less than several hundred meters (depending on frequency) resulting in substantially higher noise levels at shallow receivers. On the basis of this and previous work there seems little doubt that this mechanism is a significant source of noise in the ocean, usually dominant from about 0.2 to 5 Hz.