Working memory performance has been inconsistently reported in autism spectrum disorders (ASD). Several studies in ASD have found normal performance in digit span and poor performance in digit symbol task although these are closely related with working memory. It is assumed that poor performance in digit symbol could be explained by confirmatory behavior, which is induced due to the vague memory representation of number-symbol association. Therefore it was hypothesized that the performance of working memory task, in which vagueness did not cause confirmatory behavior, would be normal in ASD. For this purpose, the Advanced Trail Making Test (ATMT) was used. The performance of digit span, digit symbol and ATMT was compared between ASD and normal control. The digit span, digit symbol and ATMT was given to 16 ASD subjects and 28 IQ-, age- and sex-matched control subjects. The scores of these tasks were compared. A significantly lower score for ASD was found only in digit symbol compared with control subjects. There were no significant difference in digit span and working memory estimated by ATMT. Discrepancy of scores among working memory-related tasks was demonstrated in ASD. Poor digit symbol performance, normal digit span and normal working memory in ATMT implied that ASD subjects would be intact in working memory itself, and that superficial working memory dysfunction might be observed due to confirmatory behavior in digit symbol. Therefore, to evaluate working memory in ASD, tasks that could stimulate psychopathology specific to ASD should be avoided.