Starting with the premise that the construct-oriented approach is the only viable approach to personality assessment, this paper considers five propositions. First, a prerequisite to generalizable and valid psychometric measurement of personality rests on the choice of broad-based constructs with systematic univocal definitions. Next, measures will be likely to demonstrate convergent and discriminant validity and relative freedom from desirable bias if, and only if, these requirements are incorporated into test construction. Third, untrained item writers produce more valid personality scales than do empirical procedures using extant pools. Fourth, convergent and discriminant validity of personality measures is demonstrable even in the face of confounded method variance, and last, a potentially fruitful avenue for further investigation is the demonstration of a number of reproducible sets of psychopathological types. Evidence supporting each of the above propositions is discussed. Some areas of research needing attention are outlined. (Author)