The storm surge and resulting flooding can cause severe damage to coastal installations such as oil refineries and nuclear power plants and can additionally pose danger to shipping activities. Prediction of surge water levels is essential to protect lives and properties and also to issue warnings during a storm period. Occurrence of a storm surge is highly uncertain in space as well as time. The increase in sea levels due to storm surge depends on factors like atmospheric pressure, wind speed and its direction, in addition to local effects in a shallower region, such as water depth, local bathymetry and shoreline geometry. In order to evaluate and predict the storm surge numerical models are popularly employed. These models are based on hydrodynamic equations governing the water motion in sea. The solution of numerical methods is achieved through finite difference or finite element schemes. Complementary to such numerical techniques this paper presents an application of two artificial intelligence schemes namely: Neural Networks (NN) and Genetic Programming (GP). Neural networks (NN) have now become an established computing tool for many applications in ocean engineering (Jain and Deo, 2006). Neural networks do not assume any mathematical model a priori and hence are more flexible in data mining. Lack of requirement of process knowledge and that of any exogenous information, data error tolerance and easy adaptability to new observations are some of the additional attributes that this scheme possesses. Like genetic algorithm (GA) the concept of genetic programming also follows the principle of ‘survival of the fittest' borrowed from the process of evolution occurring in nature. But unlike GA its solution is a computer program or an equation as against a set of numbers in the GA and hence it is convenient to use the same as a regression tool rather than an optimization one like the GA. A good explanation of various concepts related to GP can be found in Koza (1992). Rao and Mandal (2005) and Lee (2006) have reported the use of NN to storm surge evaluation; however their works need to be more generalized by the use of a larger database and by modeling higher surge activity. There is also a need to see how NN is useful to predict the surge on real time basis. There are no reported applications of GP in the problem domain. The current study addresses these issues. It aims at obtaining water levels raised by a storm surge based on the observed time history of such levels at a given location and alternatively considering the causative factors. While the former enables real time forecasting the latter produces simultaneous estimates. In order to carry out this study storm water levels generated due to hurricanes in the Gulf of Mexico, during the years 2003 to 2005 (Fig.1) were considered. The major hurricanes namely: Ivan, Dennis, Katrina, Rita were included in the study for model training and testing. The surge prediction was made over the locations 42001, 42035 and 42039 (Fig.1) representing deep, near shore and shallow water conditions respectively in the Gulf of Mexico. In case of the real time forecasting exercise only two preceding observations were found to be sufficient to impart training to the models. The lead time considered was 1, 2, 3, 4, 6, 9 and 12 hr. A three layered NN of feed forward type and GP models were built using the first segment of 70% data and tested on remaining 30% data. Typically for Station 42001 the major hurricane events considered for the training were Dennis, Early, Katrina, Rita while those for testing included Emily, Ivan, Jenny. A typical testing performance for station 42001 in terms of the time series plot over a lead time of 6hr is shown in Figure 2. The testing performance of the GP models was found to be better than that of the NN models as reflected in high correlation coefficient (R) and low mean absolute error (MAE) and low root mean square error (RMSE). The performance of NN models was found to deteriorate strongly with increasing lead time, unlike that of the GP. As regards the causal model is concerned the major factors influencing the surge levels, namely, wind speed, wind direction, wind gust and barometric pressures were given as input to both NN and GP models. First a sensitivity analysis of these input variables was carried out to confirm their necessity in modeling. The predictions were carried out by considering the hurricane events occurring at the selected stations from 2003 to 2005. Fig. 3 shows the testing outcome in terms of the predicted time history of the storm water level in training as well as testing of the GP model. Although a satisfactory prediction of the storm water levels was obtained with the NN model (R= 0.89, RMSE = 0.51m and MAE=0.32 m) the GP outcome showed relatively better performance (R=0.90, RMSE=0.23m and MAE = 0.19 m). The present study indicates viability of the AI techniques of NN as well as GP in predicting the storm surge and may inspire more applications in this regard in future.