One of the goals of time-lapse seismic monitoring is the direct detection of the fluid front and two-phase contact area. However, several factors affect the quality of time-lapse seismic difference data and decrease detectability. One of these factors is random noise. In this paper, we propose five different methods aimed at improving the quality and detectability of noisy time-lapse seismic difference data. Common to these methods is the transform of the differences to a domain where the time-lapse signal and random noise are well separated. Our proposed methods include direct Fourier transform based spectral decomposition, bispectra, wavelet transform, singular value decomposition and hybrid methods. We also propose a method that combines multiple time-lapse difference data and gives a final difference which enhances the common part and attenuates the differences of the multiple difference images resulting in a better detectability than the original images. A synthetic time-lapse model is used to demonstrate the feasibility of our proposed methods.