Most of the existing rule induction algorithms are computationally cumbersome and consume much time specially on large noisy databases. This paper presents an efficient algorithm for extracting accurate and comprehensible set of rules from database. The algorithm transfers the attribute of continuous values into linguistic terms using suitable membership function (fuzzification). This transformation leads to the reduction of search space. The algorithm controls the rules induction through three levels. These levels are the search level (level1), the confidence level (pc), and the support level (DBC)