Two hybrid coupled models (HCMs), an intermediate complexity dynamical ocean model coupled to either a nonlinear neural network atmosphere (NHCM) or a linear regression atmosphere (LHCM), have been developed for the tropical Pacific. The ENSO (El Niño-Southern Oscillation) characteristics of the two coupled models were investigated. The results show that the NHCM can produce more realistic ENSO oscillatory behavior, with a period of about 57 months in comparison with a period of 87 months in the LHCM. With the gradual increase of coupling strength, both NHCM and LHCM exhibit phase-locking, eventually becoming a biennial oscillation with ENSO peaks in winter, indicating that the seasonal cycle is important in the low-frequency oscillations of both coupled models. The NHCM phase-locking is more realistically distributed among the calendar months, in the contrast to the very narrow phase-locking of the LHCM. Sensitivity experiments show that in the absence of external forcing, neither NHCM nor LHCM displays the irregular behavior of ENSO oscillations, suggesting that nonlinear chaotic behavior might not play a central role in ENSO oscillations, and stochastic forcing is likely to produce the irregularity of ENSO oscillations.