The accuracy in measurements of mono- and biexponential transverse relaxation processes with an MRI unit (0.5 T) was studied with a binary phantom. Comparison with spectrometer measurements (0.5 T) demonstrated that the imager underestimated the T2 values for monoexponential processes. Numerical resolution of biexponential processes also yielded underestimated relaxation times, but the resolution of a slow, constant component from faster components was relatively precise and consistent, provided the T2 ratio was above 2.5 in the T2 range 200-800 ms for a spin-echo sequence with 32 echoes. The effects of signal-averaging, strength of slice-selective gradient, single- versus multi-slice mode and repetition time were of little importance. In coronal slices a spurious biexponentiality occurred occasionally from monoexponential sources. The influence of stochastic noise was of minor importance compared to the effect of systematic noise.