Spatio-temporal variation of earthquake activity is modeled assuming fluid migration in a narrow porous fault zone whose boundaries are impermeable. The duration of earthquake sequence is assumed to be much shorter than the recurrence period of characteristic events on the fault. Principle of the effective stress coupled to the Coulomb failure criterion introduces mechanical coupling between fault slip and pore fluid pressure. A linear relation is assumed in our simulations between the accumulated slip and fault zone width on the basis of laboratory and field observations. High complexity is observed in the rupture activity so long as an inhomogeneity is introduced in the spatial distribution of initial strength, which is defined as the fracture threshold stress before the intrusion of the fluid. Frequency-magnitude statistics of intermediate-size events obeys the Gutenberg-Richter relation for all the models in which spatial heterogeneity is introduced for the initial strength. The behavior of larger-size events seems to be rather model dependent. It is also observed that the rupture occurrence tends to be inactivated immediately before the occurrence of the largest event in a sequence. This never happens if a brittle rupture is assumed in an elastic medium with no mechanical effect of fluid. This inactivation will occur because it takes much time to build up fluid pressure to break a fault segment having high initial strength, whose rupture triggers the largest event in a sequence. Our calculations also show that a single predominant principal event cannot be observed in a sequence when both the variance and average value of the distributed initial strengths are large. This may explain a feature observed for earthquake swarm.¿ 1997 American Geophysical Union