Pigeons were placed in a long chamber equipped with one key and feeder at each end side and one key and houselight at the middle. To obtain food the birds had to choose one side key after a short signal and the other side key after a long signal. The signals consisted of the illumination of the center key and the houselight and were initiated by a peck at the center key. The chamber had sensitive floor panels that enabled us to measure the location of the bird during the signals. In Experiment 1, after the birds learned the discrimination we reversed the assignment of keys to signals. In Experiment 2, we examined performance on two pairs of discriminations holding the same ratio. In Experiment 3, after the pigeons learned to discriminate two signals, we changed the duration of the long signal. The results showed that (a) the birds' motion during the signal was highly stereotypical, i.e. the birds moved to the short side, waited a few seconds, and then departed to, and stayed on the long side; (b) this motion pattern predicted the results of generalization tests with novel durations; (c) the mean of the times of departure from the short side approached its steady state values quicker than the standard deviation and consequently superposition of behavioral measures became stronger with training; (d) only the duration of the short signal influenced significantly the moment the birds departed from the short side; finally (e) the times of arrival at and departure from the short side were positively correlated, but the times of arrival and residence at the short side were negatively correlated.