In this paper, I suggest that the disjunctive feature values that originate from complicated Korean morphologies be minimized or, more radically, be eliminated, with a view to improving the efficiency and performance of natural language processing. Traditional feature descriptions rely on parts of speech and grammatically significant inflections, which are used as primary constraints in the construction of grammars. But large sets of relatively unconstrained features are likely to result in unification failure or inefficient processing because parts of speech and inflections are liable to conflict in several ways and for Korean morphology many disjunctive feature structures will have to be introduced for grammatical constraints. To resolve this problem, I introduce two well-typed structures, POS_INFL and GF_INF, which are obtained by merging part of speech and hrammatical functions with inflections. This approach allows us to eliminate a significant number of disjunctive feature values in the description, thus enabling efficient processing of feature structures. Apart from efficiency considerations, the approach can also benefit grammar engineers who have to come up with exclusive constraints when writing precise but comprehensive grammars.