Mazzotti and Adams (2004) estimated that rapid deep slip during typically two week long episodes beneath northern Washington and southern British Columbia increases the probability of a great Cascadia earthquake by 30–100 times relative to the probability during the ∼58 weeks between slip events. Because the corresponding absolute probability remains very low at ∼0.03% per week, their conclusion is that though it is more likely that a great earthquake will occur during a rapid slip event than during other times, a great earthquake is unlikely to occur during any particular rapid slip event. This previous estimate used a failure model in which great earthquakes initiate instantaneously at a stress threshold. We refine the estimate, assuming a delayed failure model that is based on laboratory‐observed earthquake initiation. Laboratory tests show that failure of intact rock in shear and the onset of rapid slip on pre‐existing faults do not occur at a threshold stress. Instead, slip onset is gradual and shows a damped response to stress and loading rate changes. The characteristic time of failure depends on loading rate and effective normal stress. Using this model, the probability enhancement during the period of rapid slip in Cascadia is negligible (<10%) for effective normal stresses of 10 MPa or more and only increases by 1.5 times for an effective normal stress of 1 MPa. We present arguments that the hypocentral effective normal stress exceeds 1 MPa. In addition, the probability enhancement due to rapid slip extends into the interevent period. With this delayed failure model for effective normal stresses greater than or equal to 50 kPa, it is more likely that a great earthquake will occur between the periods of rapid deep slip than during them. Our conclusion is that great earthquake occurrence is not significantly enhanced by episodic deep slip events.