Pigeons were trained in a matching task in which samples involved presentation of a white line on a green background (feature-present) or on an otherwise dark key (feature-absent). After asymmetrical training in which one group was initially trained with the feature-present sample and another was initially trained with the feature-absent sample, marked retention asymmetries were obtained. In both groups, accuracy dropped precipitously on trials involving the initially trained sample and remained high on trials involving the sample introduced second in training. It was concluded that asymmetrical training encouraged a single-code/default strategy in which only the sample trained initially was coded. Consistent with this conclusion, changing attributes of either sample reduced accuracy to a greater extent in pigeons initially trained with that sample than in pigeons for which that sample was introduced second in training.