Brain slices serve as useful models for the investigation of epilepsy. However, the preparation of brain slices disrupts circuitry and severs axons, thus complicating efforts to relate epileptiform activity in vitro to seizure activity in vivo. This issue is relevant to studies in transverse slices of the piriform cortex (PC), the preparation of which disrupts extensive rostrocaudal fiber systems. In these slices, epileptiform discharges propagate slowly and in a wavelike manner, whereas such discharges in vivo propagate more rapidly and jump abruptly between layers. The objective of the present study was to identify fiber systems responsible for these differences. PC slices were prepared by cutting along three different nearly orthogonal planes (transverse, parasagittal, and longitudinal), and epileptiform discharges were imaged with a voltage-sensitive fluorescent dye. Interictal-like epileptiform activity was enabled by either a kindling-like induction process or disinhibition with bicuculline. The pattern of discharge onset was very similar in slices cut in different planes. As described previously in transverse PC slices, discharges were initiated in the endopiriform nucleus (En) and adjoining regions in a two-stage process, starting with low-amplitude "plateau activity" at one site and leading to an accelerating depolarization and discharge onset at another nearby site. The similar pattern of onset in slices of various orientations indicates that the local circuitry and neuronal properties in and around the En, rather than long-range fibers, assume dominant roles in the initiation of epileptiform activity. Subtle variations in the onset site indicate that interneurons can fine tune the site of discharge onset. In contrast to the mode of onset, discharge propagation showed striking variations. In longitudinal slices, where rostrocaudal association fibers are best preserved, discharge propagation resembled in vivo seizure activity in the following respects: propagation was as rapid as in vivo and about two to three times faster than in other slices; discharges jumped abruptly between the En and PC; and discharges had large amplitudes in superficial layers of the PC. Cuts in longitudinal slices that partially separated the PC from the En eliminated these unique features. These results help clarify why epileptiform activity differs between in vitro and in vivo experiments and suggest that rostrocaudal pyramidal cell association fibers play a major role in the propagation of discharges in the intact brain. The longitudinal PC slice, which best preserves these fibers, is ideally suited for the study their role.