We tested a California sea lion for visual oddity learning by presenting problems composed of three two-dimensional black-and-white stimuli, two identical (S−) and one different (S+). In the first experimental stage, a single problem per session was presented until learning criterion was reached. In the second experimental stage, all problems were presented only five times in succession; then a new problem was introduced (six problems/session). In the third experimental stage, each problem was presented only once. The sea lion mastered all stages of oddity learning. A final transfer test with oddity problems composed of completely new stimuli yielded performance significantly above chance. Data analyses suggested learning of specific stimulus properties in the first stage, learning set formation in the second stage, but oddity conceptualization in the third stage.