Data collected in a study of hospital staff attitudes to drug addicts and other types of patient are used to illustrate the problem of concept–scale interaction in semantic differential research. Results of factor analysis of the four within-concepts correlation matrices diverged sharply from the results of the factor analysis of the across-concepts correlation matrix. Although the first two unrotated orthogonal factors could be identified in the ratings of all concepts, some of the rotated factors did not appear for some concepts. For the purposes of comparing the concepts in terms of the factors, scores were computed for each factor from the across-concepts factor-score estimation weights, but only for those concepts to which the factors applied. It is concluded that this procedure has some advantages over other methods for overcoming the problem of concept–scale interaction which have been proposed in the past.