A numerical hindcast of the Pacific Ocean circulation for the period of 1988--1998 is conducted to study the dynamics of interannual variations in the 1990s. The Poseidon quasi-isopycnal ocean model is forced with an SSMI-derived wind stress product and sea surface heat flux from an atmospheric mixed layer model. On the basis of a good comparison of the hindcast results with TAO mooring observations, the roles of the propagation and reflection of the equatorial Kelvin and Rossby waves in the evolution of El Ni¿o and La Ni¿a events in the 1990s are investigated. The linear waves are extracted from the nonlinear model in such a way as to preserve model nonlinearity so that the validity of linear theory of the wave propagation and reflection at the western and eastern boundaries can be examined. It is found that the reflection of the Kelvin waves at the eastern boundary is in very good agreement with a linear reflection at a straight meridional wall. In contrast, the reflection of the Rossby waves at the Pacific western boundary is significantly different from linear theory. Besides the validation of linear theory, the decomposition also shows marked differences from altimetry data analyses. Since the hindcast solution is dynamically consistent and close to observations, a significant part of the differences is attributed to errors associated with the incomplete (least squares fit) wave decomposition from the altimetry data. On the basis of the wave decomposition, the dynamics of the simulated El Ni¿o and La Ni¿a events in the 1990s are compared with three ENSO paradigms. The investigation shows that precursive Rossby waves, important in the Delayed Oscillator paradigm, played an important role in the termination of the 1991--1993 and the 1997--1998 El Ni¿o events. Reflected Rossby waves from the Pacific eastern boundary, the primary negative feedback in the Advective-Reflective paradigm, were present during the termination of the 1994--1995 and the 1997--1998 El Ni¿o. The Western Pacific Oscillator is at best at work during the decay of the 1997--1998 El Ni¿o; however, it played a minor role compared with the other paradigms. None of the three paradigms appears to describe the termination of the 1988--1989 and the 1995--1996 La Ni¿a events and the onset of the 1997--1998 El Ni¿o event well.