Usually, supervised learning systems utilize only the information provided by the prelabeled training data and rules are generated without looking at the testing examples. After learning, the testing data are used to evaluate the performance of the learning system. When the system later encounters a testing example which is not covered by any existing rules, this example will be classified according to some default method, which may not be very informative. In this case, the information carried by this testing example is somehow ``wasted''. When none of the learned rules matches a test observation, it would be desirable to learn a new rule regarding the particular uncovered observation and classify the observation using this new rule. Intuitively, we might learn this new rule covering the test item by applying a BruteDL-like search over the training data, which we have retained. However, a BruteDL search may be too costly at classification time. Thus, we use a greedy search procedure adapted from the OSR (Optimized Set Reduction) [1] system. In particular, we use a default strategy, Minimum OSR, to classify test objects not covered by rules learned by BruteDL.