A Japanese version of the Autism Spectrum Quotient (AQ), AQ-J was administered to 25 normally intelligent high-functioning pervasive developmental disorder (HPDD) patients (mean age, 24.2 years; 24 male, one female) and 215 controls (mean age, 30.4 years; 86 male, 129 female) randomly selected from the general population. The AQ-J had satisfactory internal consistency reliability (Cronbach's alpha > 0.70 in the two groups), test-retest reliability, and discriminant validity [i.e. the AQ-J score was significantly higher in the HPDD (mean, 29.6) than controls (mean, 22.2)]. At a cut-off of 26, the AQ-J had satisfactory sensitivity, specificity, and negative predictive value, but it had low positive predictive value (0.24) possibly due to the facts that the 25 mild HPDD patients scored lower and the controls scored higher on the AQ-J than British counterparts on the AQ. The AQ-J-21 (consisting of 21 items significantly associated with HPDD diagnosis) and the AQ-J-10 (consisting of 10 of the 21 items with an effect size > 0.17) had higher, although not satisfactory, positive predictive values of 0.35 and 0.46 at cut-offs of 12 and 7, respectively, than the AQ-J. The AQ-J and two short forms are useful not to predict but to rule out mild HPDD, the most difficult part of HPDD to be distinguished from non-PDD conditions, in persons scoring under the cut-offs and to consider professionals' examination of HPDD in persons scoring over them, because their negative predictive values were satisfactory.