Quantitative processing of NMR relaxation images depends on the characteristics of the used fitting algorithm. Therefore several common fitting algorithms are compared for decay curves with low signal-to-noise ratios. The use of magnitude data yields a non-zero base line, and is shown to result in an overestimation of the decay time. A simple base line correction is no solution since this yields an equally large underestimation due to overcorrection of the first part of the curve. The use of squared data does yield reliable results, but only in the case of monoexponential decays. The best fitting algorithm under all experimentally occurring conditions turns out to be using real data after phase correction. A phase correction scheme is proposed, which applies to all imaging experiments for which the phase of the pixels is constant over the echo train. This scheme is validated for a phantom and for a tulip bulb.