The present experiments demonstrated that, in the rabbit nictitating membrane preparation, a conditioned response (CR) can be selectively eliminated in one portion of a conditioned stimulus (CS) while it is still paired with the unconditioned stimulus (US). Rabbits were initially trained with two stimuli (tone, light). Each was paired with the US by using a mixture of two CS-US interstimulus intervals (ISIs): 200 ms and 1,200 ms in Experiment 1; 150 ms and 500 ms in Experiment 2. The CRs showed double peaks, one for each ISI. Subsequently, one CS (A) was trained with only the longer ISI, whereas the other CS (B) continued to be trained with both ISIs. Consequently, the CR peak based on the shorter ISI disappeared for CSA but not for CSB. The later CR peaks during both CSA and CSB were maintained. These results support time-based models of conditioning. Implications for proposed mechanisms of extinction are discussed.