following 20 sessions of variable-interval 20-sec reinforcement in the presence of a single 45-deg line-tilt stimulus, three pigeons were trained to discriminate between line tilts of 45 deg correlated with variable-interval 2Q..sec reinforcement and line tilts of 15 deg correlated with extinction. A generalization test along the line-tilt dimension was administered following a criterion discrimination performance. Gradients derived in terms of relative frequency of response as a function of line tilt indieated strong externalszimulus control and exhibited clear peak shift. From the interresponse time (lRI) distributions generated for responding to each test stimulus, probability of response conditional upon IRI (lRIs/Op) was derived as a joint function of line tilt and IRI. The IRIs/Op functions for responses following IRIs in 0.2-sec-wide classes from 0.2 to 1.0 sec and for responses following IRIs in the interval of 1.0 to 2.0 sec were similar to the relative generalization gradients and also exhibited peak shift. Few IRIs were greater than 2.0 sec. External stimulus control was established over responses terminating IRIs both longer and shorter than 1.0 sec. .Stimulus control is assessed by measuring the change in response probability that results from a change in stimulus value. The generalization gradient obtained following free-operant, single-stimulus training (Guttman & Kalish, 1956) or discrimination training (Hanson, 1959) shows variation in rate or frequency of response. In the generalization test, response frequency is recorded in each test period (which is typically 30 sec in duration), and responses to the test stimulus are summed over all test periods. The problem with this particular method is that it gives no information about changes in rate of response that might occur within a test period. That is, although there may be variation in interresponse time (IRT), the defining characteristic of rate, such variation cannot be observed without disecting the gross frequency measure. If there is variation in IRT within the test periods, it is possible that generalization is different for different classes of IRT. In other words, the exteroceptive stimuli control responses which occur after some IRTs, but do not control those responses which occur after other IRTs. Blough (1963) presents data that suggest that stimulus generalization is different for different IRT classes. The IRT distributions for a pigeon's responses to each of several test wavelength stimuli (obtained following training with one stimulus value) showed that responses terminating short IRTs are the least affected by stimulus value. Blough (1963) concludes that "the gradient is produced mainly by a shift upward of the longer IRTs which leave the time of the short peaks relatively unaffected. It appears, as might be expected, that stimulus control is exerted over the responses following long IRTs: the thing that controls the occurrence of a *This research was conducted at the University of Otago.