Signal separation is one of the significant topics in the signal processing era. How to obtain the separated signal from the mixed (source) signals is the key issue. In general, the statistical characteristics are utilized in analysing the signals. In this study, the mixed two-dimensional (2-D) signals with Gaussian distributions are consider in the signal separation task. The exponential type of linear frequency modulation (LFM) signals has the feature of energy concentration in certain fractional order of the fractional Fourier transform (FrFT). By searching the highest energy distribution, one of the mixed signals can be separated and can be reconstructed using the inverse FrFT. The experimental results show that all signals are successfully separated and can be reconstructed with high fidelity. The correlation coefficients between the original and the reconstructed signals can reach as high as 0.9998.