In the Lamont coupled model of the tropical Pacific, the ocean model uses a simple parameterization scheme for the subsurface temperature, which is replaced here by a neural network (NN) scheme. The inputs of the NNs are the six leading principal components (PCs) (i.e. EOF time series) of the 20^oC isotherm depth, and the output is one of the five leading PCs of the subsurface temperature anomalies. Forced by the Florida State University wind stress, the ocean models were run from January 1964 to August 2001. The NN-enhanced Lamont ocean model simulated the sea surface temperature anomalies (SSTA) better than the original Lamont ocean model in the central and western equatorial Pacific. Improvements by the NN-enhanced model is also seen from the principal component analysis (PCA) and nonlinear principal component analysis (NLPCA) of the SSTA.