The behavior of noise in asynchronous spectrum in generalized two-dimensional (2D) correlation spectroscopy is investigated. Mathematical analysis on the noise of 2D spectra and computer simulation on a model system show that the fluctuation of noise in a 2D asynchronous spectrum can be characterized by the standard deviation of noise in 1D spectra. Furthermore, a new approach to improve a signal-to-noise ratio of 2D asynchronous spectrum by a Butterworth filter is developed. A strategy to determine the optimal conditions is proposed. Computer simulation on a model system indicates that the noise of 2D asynchronous spectrum can be significantly suppressed using the Butterworth filtering. In addition, we have tested the approach to a real chemical system where interaction between berberine and β-cyclodextrin is investigated using 2D UV-vis asynchronous spectra. When artificial noise is added, cross peaks that reflect intermolecular interaction between berberine and β-cyclodextrin are completely masked by noise. After the method described in this article is utilized, noise is effectively suppressed, and cross peaks are faithfully recovered. The above result demonstrates that the approach described in this article is applicable in real chemical systems.