A new theoretical framework for the evaluation of the in-band nonlinear distortion effects on the performance of OFDM systems is presented. In contrast to previous works that approximate the nonlinear noise as a Gaussian additive random process, the new framework is based on the properties of the large deviations of a stationary Gaussian process and shot noise theories, which can evaluate the performance of the OFDM system with high accuracy, especially at realistic scenarios where the Gaussian approximation of the nonlinear noise is no longer valid. The approach can be used to evaluate many communication systems with peak-limited nonlinearities and high PAPR, such as the downlink performance analysis of large capacity DS-CDMA systems.OFDM, CDMA, nonlinearities, impulse noise OFDM is a promising candidate for achieving high data rate transmission in wireless environment and is widely employed in communication systems. Since an OFDM signal is the sum of several statistically independent random subcarriers, its baseband in-phase/quadrature (I/Q) components can be approximately represented as a Gaussian process with Rayleigh envelope distribution and uniform phase distribution invoking the Central Limit Theorem [ ] 1 . Simulation results show that the Gaussian approximation is very realistic for most of the practical systems of interest with a sufficiently large number of subcarriers. One of the major drawbacks of a nearly Gaussian OFDM waveform is a greatly variable envelope or high peak-to-average power ratio (PAPR) that makes it particularly sensitive to nonlinear distortions, such as analog-to-digital (A/D) converters, IFFT/FFT processors with finite word length, RF high power amplifiers (HPA), etc., which will cause both in-band and out-of-band distortion. It has been shown that the out-of-band distortion can be removed using a simple filter without adding any distortion to the signal [ ] 2 . Recently, several different methods have been derived for analysis of in-band distortion of OFDM systems. The effect of the distortion can be modeled as attenuation of the signal and generation of a nonlinear additive noise [ ] 3 . This nonlinear noise