The relation between performance during discriminative training and subsequently obtained measures of stimulus control was investigated. Pigeons served as experilmental subjects. In the discriminative training phase, a single peck on the center key, transilluminated by a bright or dim white light, resulted in the onset of the side keys, one red and one green. If the center key was brightly lighted, a response on the red side key was correct. A response on the green side key vas correct if the center key was dimly lighted. Correct responses were reinforced on independently arranged variable-interval schedules. Following discriminative training, tests of stimulus control vere administered during which white light of 11 intensities was projected on the center key and responses on the red and green side keys recorded. The proportion of correct responses in the presence of a bright or dim center-key stimulus decreased with decreases in the frequency of reinforcement of correct red or correct green responses, respectively. The slopes of the stimulus control gradients were related to the extent of response bias during training. The greater the bias to respond on the green key, the flatter the gradient showing the proportion of green-key responses to each stimulus and the steeper the corresponding gradient of red-key responses.The control of behavior by exteroceptive stimuli is commonly assessed by comparing response rates in the presence of each stimulus from a set of stimuli differing along a single dimension. If the response rates do not differ in the presence of different stimuli, poor stimulus control is inferred. If the rate of responding is high in the presence of one stimulus, and consistently decreases as stimuli increasingly different from that stimulus are presented, a higher degree of stimulus control is inferred. Much criticism has been levelled at the use of such response rate measures to assess stimulus control (Blough, 1963; Ray and Sidman, 1970;Jenkins, 1965). In the first place, rate of responding may be controlled by factors other than stimulus value, thus masking changes in response probability correlated with changes in stimulus value. A related line of criticism has been developed by several authors, who have suggested that the behav1This paper is based on a dissertation submitted to