Susceptibility-induced geometry and intensity distortions are a familiar observation in MR imaging. In the past few years several attempts have been made to aid in the understanding of susceptibility artifacts by means of simulation studies. Although these studies, which were mostly carried out with simple test objects, have produced some qualitative insight into chi-artifacts, the results lacked precision in describing finer details. In this paper we show the discrepancy between theory and experiment in previous work to be the result of an inadequate theoretical approach. In most studies so far, delta B0 effects are taken into account in the frequency domain, that is, after Fourier transformation of the data. In our view the simulation should follow the actual sequence of events in an imaging experiment and deal with the effect of error fields in the time domain (k-space) already. The correctness of this view is demonstrated here by comparing the results of time and frequency domain simulation against experimental observation for a coaxial cylinder phantom, a widely used model in this type of work. Having established the superiority of the time domain simulation, we demonstrate its use in predicting chi-artifacts under various experimental conditions, for example, in spin-echo and gradient-echo imaging with a reduced number of phase-encoding steps.