Background: Working memory (WM) is often assessed with serial order tests such as repeating digits backward. In prior dementia research using the Backward Digit Span subtest (BDT) only aggregate test performance was examined. The current research tallied primacy/ recency effects; out-of-sequence transposition errors; perseverations and omissions to assess WM deficits in patients with mild cognitive impairment (MCI). Methods: Memory clinic patients (n= 66) were classified into three groups -- single domain amnestic MCI (aMCI), combined mixed domain/ dysexecutive MCI (mixed/dys MCI), and non-MCI where patients did not meet criteria for MCI. Serial order/ WM ability was assessed by asking participants to repeat 7 trials of five digits backwards. Serial order position accuracy, transposition errors, perseverations, and omission errors were tallied. Results: A 3 (group) x 5 (serial position) repeated measures ANOVA yielded a significant group x trial interaction. Follow-up analyses found an absence of a recency effect for mixed/dys MCI patients. Mixed/dys MCI patients scored lower than non-MCI patients for serial position 3 (p< 0.003) serial position 4 (p< 0.002); and lower than both group for serial position 5 (recency; p< 0.002). Mixed/dys MCI patients produced more transposition errors than both groups (p< 0.010); and more omissions (p< 0.020), and perseverations errors (p< 0.018) than non-MCI patients. Conclusions: The striking absence and/ or attenuation of a recency effect using serial order parameters obtained from the BDT may constitute a neurocognitive biomarker for WM deficits in MCI and provide additional diagnostic information regarding working memory deficits in MCI.