We describe the serial implicit learning task (SILT), a novel test for assessing implicit and procedural learning in rodents, and have used the task to test whether striatal lesions disrupt the speed and accuracy of responding to stimulus-response (S-R) sequences that may be either predictable or unpredictable. In this task, the rats must learn to respond to two consecutive stimulus lights, S1 and S2, which may occur on each trial in any of five alternative response locations, for food reward. For two of the S1 locations, the location of the subsequent S2 is predictable, whereas for the other three S1 locations, the location of S2 is unpredictable and can appear with equal probability in any of the other four locations (i.e. any open hole other than S1 itself). All rats learned to make the serial responses rapidly and accurately. Effective implicit learning was demonstrated by there being a significant advantage in both speed and accuracy in responding to predictable than to unpredictable S2 stimuli. Following quinolinic acid lesions of the medial or lateral striatum, the lesioned rats showed significantly reduced accuracy and increased latencies in responding to both S1 and S2, although (contrary to initial hypothesis) the benefits of predictability of S2 were retained. These data establish a novel and efficient operant test for implicit learning in the rat.