An experimental paradigm of contingent classification is introduced, in which stimuli in a classification task are cued by the addition of another dimension so that the classification is reduced to the discrimination of just two stimuli. The contingencies are arranged so that the cuing dimension provides no response information. Equivalent contingencies can be arranged by precuing in advance of the stimulus. Four experiments explored this paradigm with a discrete reaction task, using stimuli C, O, L, and T. In Experiment 1, with no additional cuing dimension, it was established that classification and discrimination based on the curved-straight distinction are easier than those not permitting its use. In Experiment 2, color was added as a cuing dimension, with both easy and hard classification tasks, in the latter case using contingencies that produced easy or hard discriminations. There was no improvement in performance with any of the contingent classifications. In Experiment 3, lateral location was used as a cuing dimension with a hard classification task, and in this case performance improved with a contingency that produced easy discriminations, but performance was not nearly as good as with the equivalent discrimination tasks alone. In Experiment 4, precuing of 500 msec was used with a hard classification task, and performance improved almost to the level of the equivalent tasks, for both easy and hard discriminations. It is suggested that the use or nonuse of the contingent relations is best explained as due to the use of optional attentional allocation.