The overriding aim of the present study was to psychometrically evaluate the Karolinska Scales of Personality (KSP) in a delinquent sample. The KSP was administered to a group of male juvenile delinquents (N = 47, mean age 17 years; SD = 1.2) from four representative Swedish national correctional institutions for serious offenders with conduct disorder, and evaluated by means of statistical methods. As expected, reliability in terms of internal consistency, measured by Cronbach's alpha, for some basic scales (Socialization, Muscular Tension, Somatic Anxiety, and Impulsiveness) was high (> .70). However, according to previous research, all aggressiveness and hostility-related scales, with the exception of Verbal aggression scale, showed low reliability. Reliability in terms of homogeneity, measured by mean interitem correlations, was acceptable (> .20) for 6 out of 15 scales. Factor analysis (maximum likelihood) arrived in this sample at a four-factor solution. Issues regardingthe interpretation of reliability, in terms of Cronbach's alpha or scale homogeneity, are highlighted. The concepts of reliability of personality measures in deviant groups are discussed and also possible guidelines for the construction of personality inventories in groups of this kind. The results are  briefly discussed from a cultural perspective.