Like the numbers in a sudoku puzzle, a lexeme’s principal parts provide enough information–but only enough–to deduce all of the remaining forms in its paradigm. Because principal parts are a distillation of the implicative relations that exist among the members of a lexeme’s paradigm, they afford an important (but heretofore neglected) basis for typological classification. We recognize three logically distinct sorts of principal-part systems that might be postulated for a given language: static, adaptive, and dynamic. Focussing for present purposes on dynamic systems, we propose five crosscutting criteria for the typological classification of principal-part systems. These criteria relate to (i) how many principal parts are needed to determine a lexeme’s paradigm; (ii) whether distinct lexemes possess parallel sets of principal parts; (iii) how many principal parts are needed to determine a given word in a lexeme’s paradigm; (iv) what sort of morphological relation exists between a principal part and the forms that it is used to deduce; and (v) whether lexemes’ nonprincipal parts are inferred from their principal parts in the same way from one inflection class to another. Drawing on these criteria, we propose a novel classification of a range of typologically diverse languages.