Dynamic causal modeling (DCM) of functional magnetic resonance imaging (fMRI) data allows one to make inferences about the architecture of distributed networks in the brain, in terms of effective connectivity. fMRI data are usually acquired using echo planar imaging (EPI). EPI sequences typically acquire slices at different times over a few seconds. DCM, in its original inception, was not informed about these slice timings and assumed that all slices were acquired simultaneously. It has been shown that DCM can cope with slice timing differences of up to 1 s. However, many fMRI studies employ a repetition time (TR) of 3 to 5 s, which precludes a straightforward DCM of these data. We show that this limitation can be overcome easily by including slice timing in the DCM. Using synthetic data we show that the extended DCM furnishes veridical posterior means, even if there are large slice-timing differences. Model comparisons show that, in general, the extended DCM out-performs the original model. We contrast the modeling of slice timing, in the context of DCM, with the less effective approach of 'slice-timing correction', prior to modeling. We apply our procedure to real data and show that slice timings are important parameters. We conclude that, generally, one should use DCM with slice timing.