Multi-exponential inversion of NMR (Nuclear Magnetic Resonance) relaxation signal has been widely used in petroleum industry both in core analysis for petrophysical studies and in logging interpretation for oil exploration. In order to obtain T2 relaxation time distributions with high resolution in the case of low SNR (Signal-to-Noise Ratio), we use simulated and experimental data to analyze the effects on the inversion of the number of pre-selected relaxation times, the number of spin echoes acquired, and the compression in time domain of the spin-echo trains. We found that reducing the number of echoes acquired can cause divergent behavior for the long T2 components, changing the number of windows for data compression does not change the shape of the inverted T2 distribution significantly. We also studied the effect of different SNRs on inverted T2 distributions for different inversion algorithms. We found that the response to different SNRs for different algorithms is different, and suggested using an adjustment parameter to correct such effects.