One of the primary results of this study is that wave group statistics observed in the ocean at depths greater than 10 m for a large range of wave conditions are consistent with linear theory. Wave group prediction models that do not include the effect of correlations between waves separated by several intervening waves underpredict the number of sequential large waves. On the other hand, a model that includes correlations of waves separated by as many as six intervening waves predicts accurately the observed wave group statistics. Another result is that although computationally convenient, wave group prediction models based on power spectral shape alone are not as accurate as models based on direct simulations of the wave field. In shallower water where nonlinearities in the wave field can be important, linear theory based models underpredict the lengths of groups of large waves, but nonlinear simulations that use information from the power spectrum and bispectrum of the time series of sea-surface elevation are more accurate. The simulations suggest nonlinearities result in longer groups of high waves than occur in a Gaussian (eg, linear) sea.