Abstract: Pigeons were trained with a conditional discrimination task in three-key operant chambers. Choosing either the left or right key was followed by food according to combinations of three preceding events: (a) a houselight illumination condition (dark or light), (b) presence or absence of green flashes on the three keys, (c) a color (amber or blue) of the center sample key. With these 2&#502&#502 event combinations, eight types of correct trials were prepared: (1) dark&#77no flash&#77amber&#77LEFT, (2) dark&#77no flash&#77blue&#77RIGHT, (3) dark&#77flash&#77amber&#77RIGHT, (4) dark&#77flash&#77blue&#77LEFT, (5) light&#77no flash&#77amber&#77RIGHT, (6) light&#77no flash&#77blue&#77LEFT, (7) light&#77flash&#77amber&#77LEFT, and (8) light&#77flash&#77blue&#77RIGHT. Seven of these eight types were used for training of a given bird, and then the remaining trial type was presented as the test. If the birds had learned the conditional structure of the events (the hierarchical switching rule), they would have responded correctly to the test type. However, they chose the opposite side key, suggesting that they had learned cue configuration or multiple rules to solve the task.