Incremental learning is motivated by a need to rapidly and frequently exploit knowledge during learning. Theoretically, it is possible to rerun inherently nonincremental methods with each new observation (plus all previous ones) and thus produce an entirely new knowledge base from scratch. Michalski (1985) has called this a revolutionary approach to knowledge base update. However, as the number and variety of observations grows, the cost of revolution becomes prohibitive. An alternative is an evolutionary strategy that changes only `faulty' parts of a knowledge base to accommodate new observations (Michalski, 1985). One disadvantage of this approach is that localized changes may result in a knowledge base of lower `quality.' Thus, incremental learning methods trade quality against cost, hopefully not to the significant detriment of the former. We now turn our attention to a number of previous studies of incremental learning with special attention to their management of the cost/quality tradeoff.