The bispectrum of several measured ocean acoustic ambient noise processes was estimated and used to test for the linearity and Gaussianity of the processes. A linear process has a constant bispectrum, while a Gaussian process has a zero bispectrum. Both linearity and Gaussianity could be easily rejected for several common ambient noise processes, even though in some cases the processes tested to be univariate Gaussian. These results suggest that autoregressive models, since they are linear models, may not be the most appropriate models for ocean acoustic ambient noise. Some possible explanations for the causes of the observed nonlinear behavior will be suggested. [Work supported by the Office of Naval Research, Code 411.]