This paper explores the use of nonlinear prediction in the modeling of sea clutter. The nonlinear methods considered here are based on local approximations: nearest neighbor and local linear prediction. The effects of radar scan ranges and the number of training samples on the nonlinear clutter model are examined. The nonlinear predictive model is then used for clutter suppression to enhance target detectability. It is shown that the nonlinear predictive detection scheme can detect small floating targets such as beach balls embedded in sea clutter. The standard linear prediction is used for comparison. It is observed that the nonlinear prediction outperforms the linear one on a regular basis.